# AI Playbook Gems > Don't just read about AI agents. Run them. Each gem is a playbook: a prompt, setup steps, a cost bucket, and a source URL only when one was confirmed. ## Pages - [Home](https://aiplaybookgems.com/): search and filters - [About](https://aiplaybookgems.com/about): verification rules and ranking method - [Submit](https://aiplaybookgems.com/submit): propose a playbook - [JSON feed](https://aiplaybookgems.com/api/usecases.json): every playbook - [RSS](https://aiplaybookgems.com/feed.xml) ## Playbooks - [Brief a topic with sources I can open](https://aiplaybookgems.com/usecases/perplexity-brief-a-topic-with-sources-i-can-open): Perplexity; Perplexity handles this as a draft: Five claims, each with the URL I gave you or the words not in the packet. Do not cite a page you did not open in the packet. - [Build a 90-day cash view from two accounts](https://aiplaybookgems.com/usecases/claude-cowork-build-a-90-day-cash-view-from-two-accounts): Claude Cowork; Claude Cowork handles this as a draft: List expected ins and outs I name. Flag weeks that go negative. Do not invent income. Do not transfer money. - [Build a competitor feature grid from pages I saved](https://aiplaybookgems.com/usecases/gemini-build-a-competitor-feature-grid-from-pages-i-saved): Gemini; Gemini handles this as a draft: Feature present or not stated. Do not guess from memory. - [Build a glossary for a new codebase](https://aiplaybookgems.com/usecases/gemini-build-a-glossary-for-a-new-codebase): Gemini; Gemini handles this as a draft: Terms from the README I paste. Do not edit the repo. - [Build a trip skeleton from constraints](https://aiplaybookgems.com/usecases/openclaw-build-a-trip-skeleton-from-constraints): OpenClaw; OpenClaw handles this as a draft: Dates, city, and the rule that I will not check a bag. Propose a day outline with no reservations. Do not book. - [Build a week of school pickup coverage](https://aiplaybookgems.com/usecases/openclaw-build-a-week-of-school-pickup-coverage): OpenClaw; OpenClaw handles this as a draft: From the constraints I list, propose who covers which day. Mark conflicts instead of silently double-booking. Do not text the family. - [Build a weekly support themes memo](https://aiplaybookgems.com/usecases/zapier-agents-build-a-weekly-support-themes-memo): Zapier Agents; Zapier Agents handles this as a draft: Five themes, each with a ticket id I can open. Do not mail the company. - [Build an account brief before a call](https://aiplaybookgems.com/usecases/grok-bot-build-an-account-brief-before-a-call): Grok Bot; Grok Bot handles this as a draft: From the pages I paste: what they sell, a recent change, and two questions. No invented metrics. Do not email the account. - [Cluster cancellation reasons](https://aiplaybookgems.com/usecases/lindy-cluster-cancellation-reasons): Lindy; Lindy handles this as a draft: From the exit notes, five reasons with counts. Do not win them back automatically. - [Compare three phone plans I already pay for](https://aiplaybookgems.com/usecases/hermes-agent-compare-three-phone-plans-i-already-pay-for): Hermes Agent; Hermes Agent handles this as a draft: Use only the plan pages I paste. Make a table of price, hotspot, and international. Mark cells you could not find. Do not switch my plan. - [Compare three product pages I already opened](https://aiplaybookgems.com/usecases/chatgpt-agent-compare-three-product-pages-i-already-opened): ChatGPT Agent; ChatGPT Agent handles this as a draft: Price, warranty, and return window in a table. Leave a cell blank rather than guessing. Do not buy. - [Create a key and code inventory](https://aiplaybookgems.com/usecases/hermes-agent-create-a-key-and-code-inventory): Hermes Agent; Hermes Agent handles this as a draft: List each lock I describe and where the spare lives. Redact nothing I did not already write. Do not store this anywhere but the file I name. - [Draft a cancellation for a free trial](https://aiplaybookgems.com/usecases/perplexity-draft-a-cancellation-for-a-free-trial): Perplexity; Perplexity handles this as a draft: The steps on the cancellation page I paste, in order, and the date the charge would hit. Do not click cancel for me. - [Draft a migration plan for one table](https://aiplaybookgems.com/usecases/devin-draft-a-migration-plan-for-one-table): Devin; Devin handles this as a draft: Columns and the backfill question. Do not run SQL. - [Explain a school device policy to a household](https://aiplaybookgems.com/usecases/claude-cowork-explain-a-school-device-policy-to-a-household): Claude Cowork; Claude Cowork handles this as a draft: What is allowed, what is not, and the consequence the policy names. Do not contact the school. - [Extract quotes from a customer interview](https://aiplaybookgems.com/usecases/lindy-extract-quotes-from-a-customer-interview): Lindy; Lindy handles this as a draft: Only sentences they said. Mark the timestamp if I included one. Do not publish. - [Extract vocabulary I marked in a text](https://aiplaybookgems.com/usecases/cursor-extract-vocabulary-i-marked-in-a-text): Cursor; Cursor handles this as a draft: Word and the sentence I highlighted. Do not translate beyond that sentence. - [Line a hospital bill up against public prices](https://aiplaybookgems.com/usecases/claude-cowork-line-a-hospital-bill-up-against-public-prices): Claude Cowork; Claude Cowork handles this as a draft: Build a spreadsheet of the line items on the bill I paste, flag charges that repeat, and list questions for the billing office. Do not call the hospital or send a dispute letter. - [List refill dates for household staples](https://aiplaybookgems.com/usecases/gemini-list-refill-dates-for-household-staples): Gemini; Gemini handles this as a draft: From the purchase dates I give, suggest the next buy date using the cadence I set. Do not subscribe. - [Make a fixture from a sanitized log](https://aiplaybookgems.com/usecases/cursor-make-a-fixture-from-a-sanitized-log): Cursor; Cursor handles this as a draft: Replace emails with placeholders. Do not use production data. - [Make a guest checklist for a spare room](https://aiplaybookgems.com/usecases/lindy-make-a-guest-checklist-for-a-spare-room): Lindy; Lindy handles this as a draft: From my notes, linens, wifi, and the one rule I want on the note. Keep it to one page. Do not text the guest. - [Make a lyrics-free jingle brief](https://aiplaybookgems.com/usecases/gemini-make-a-lyrics-free-jingle-brief): Gemini; Gemini handles this as a draft: Mood, length, and the words I will not use. Do not generate a final mix for release. - [Map entry points before a risky change](https://aiplaybookgems.com/usecases/claude-code-map-entry-points-before-a-risky-change): Claude Code; Claude Code handles this as a draft: Scripts, main files, and the test command in the README. Do not push. - [Plan an airport transfer with a time cushion](https://aiplaybookgems.com/usecases/chatgpt-agent-plan-an-airport-transfer-with-a-time-cushion): ChatGPT Agent; ChatGPT Agent handles this as a draft: Leave-by time using the durations I provide. Show the arithmetic. Do not hire a car. - [Prep an agenda for a 25-minute staff meeting](https://aiplaybookgems.com/usecases/lindy-prep-an-agenda-for-a-25-minute-staff-meeting): Lindy; Lindy handles this as a draft: Three topics from the doc, each with a decision or not. Do not send the invite. - [Prep an interview guide from a research question](https://aiplaybookgems.com/usecases/chatgpt-agent-prep-an-interview-guide-from-a-research-question): ChatGPT Agent; ChatGPT Agent handles this as a draft: Eight questions. No leading answers. Do not contact participants. - [Prep discovery questions for one industry](https://aiplaybookgems.com/usecases/zapier-agents-prep-discovery-questions-for-one-industry): Zapier Agents; Zapier Agents handles this as a draft: Eight questions tied to the industry page I paste. Do not contact the prospect. - [Prep questions before a mortgage recast call](https://aiplaybookgems.com/usecases/grok-bot-prep-questions-before-a-mortgage-recast-call): Grok Bot; Grok Bot handles this as a draft: From the letter, list the fee, the new payment, and three questions. Do not request the recast. - [Propose chapter titles for a draft](https://aiplaybookgems.com/usecases/crewai-propose-chapter-titles-for-a-draft): CrewAI; CrewAI handles this as a draft: Titles that match sections I marked. Do not rewrite the book. - [Sort a shoebox of contractor invoices](https://aiplaybookgems.com/usecases/n8n-sort-a-shoebox-of-contractor-invoices): n8n; n8n handles this as a draft: Name the vendor, date, amount, and whether the PDF includes an address. Flag invoices missing a date. Do not pay anyone. - [Sort the morning mail into four piles](https://aiplaybookgems.com/usecases/openclaw-sort-the-morning-mail-into-four-piles): OpenClaw; OpenClaw handles this as a draft: From the inbox I approve, label each thread: reply, waiting, fyi, or ignore. Draft replies only for reply. Do not send mail. - [Summarize a benefits booklet section on therapy](https://aiplaybookgems.com/usecases/hermes-agent-summarize-a-benefits-booklet-section-on-therapy): Hermes Agent; Hermes Agent handles this as a draft: Visit limits and whether a referral is mentioned. Do not book therapy. - [Summarize a caregiver update](https://aiplaybookgems.com/usecases/gemini-summarize-a-caregiver-update): Gemini; Gemini handles this as a draft: Changes since yesterday in the note I paste. Do not text the family thread. - [Summarize a student-loan letter](https://aiplaybookgems.com/usecases/claude-summarize-a-student-loan-letter): Claude; Claude handles this as a draft: Balance, payment, and the date something changes. Unknown if the letter does not say. Do not consolidate. - [Summarize access reviews due this month](https://aiplaybookgems.com/usecases/hermes-agent-summarize-access-reviews-due-this-month): Hermes Agent; Hermes Agent handles this as a draft: System, owner, and due date from the sheet. Do not revoke access. - [Summarize visa checklist items still missing](https://aiplaybookgems.com/usecases/manus-summarize-visa-checklist-items-still-missing): Manus; Manus handles this as a draft: Compare the embassy list I paste to the documents I say I have. Do not submit an application. - [Triage a support queue into four buckets](https://aiplaybookgems.com/usecases/n8n-triage-a-support-queue-into-four-buckets): n8n; n8n handles this as a draft: Billing, bug, how-to, and unknown. One sentence why. Do not reply to customers. - [Turn a chapter into ten questions I can answer](https://aiplaybookgems.com/usecases/claude-turn-a-chapter-into-ten-questions-i-can-answer): Claude; Claude handles this as a draft: Questions from the chapter I paste, with the page line they came from. Do not take the quiz for me. - [Turn a rough brief into three directions](https://aiplaybookgems.com/usecases/claude-turn-a-rough-brief-into-three-directions): Claude; Claude handles this as a draft: Each direction uses only the constraints in the brief. Do not publish. - [Turn a visit summary into questions for the next appointment](https://aiplaybookgems.com/usecases/claude-turn-a-visit-summary-into-questions-for-the-next-appointment): Claude; Claude handles this as a draft: Questions only, tied to words in the summary. No new diagnoses. Do not message the clinic. - [Write a babysitter briefing for one evening](https://aiplaybookgems.com/usecases/hermes-agent-write-a-babysitter-briefing-for-one-evening): Hermes Agent; Hermes Agent handles this as a draft: Bedtime, allergies, and the show they can watch. One page. Do not contact the sitter. - [Write a Monday status from ticket titles](https://aiplaybookgems.com/usecases/n8n-write-a-monday-status-from-ticket-titles): n8n; n8n handles this as a draft: Done, next, and blocked, using only the titles I paste. No invented progress. Do not post in Slack. - [Build a candidate shortlist without contacting anyone](https://aiplaybookgems.com/usecases/grok-bot-talent-scout): Grok Bot; A Grok Bot turns a role description into 20 sourced candidate notes and outreach drafts. The first run is not allowed to message candidates. - [Build a clinic visit folder checklist](https://aiplaybookgems.com/usecases/claude-cowork-build-a-clinic-visit-folder-checklist): Claude Cowork; Claude Cowork handles this as a draft: ID, insurance card, and the questions doc. Do not upload them anywhere. - [Build a follow-up list of stale opportunities](https://aiplaybookgems.com/usecases/manus-build-a-follow-up-list-of-stale-opportunities): Manus; Manus handles this as a draft: Name, last touch, and a draft sentence. I choose who gets it. Do not send or edit amounts. - [Build a gift budget that survives December](https://aiplaybookgems.com/usecases/hermes-agent-build-a-gift-budget-that-survives-december): Hermes Agent; Hermes Agent handles this as a draft: Spread the cap I give you across the people on my list. Show what is left. Do not buy anything. - [Build a move-out cost checklist from the lease](https://aiplaybookgems.com/usecases/grok-bot-build-a-move-out-cost-checklist-from-the-lease): Grok Bot; Grok Bot handles this as a draft: List deposits, cleaning clauses, and deadlines. Quote the clause number. Do not message the landlord. - [Categorize a month of spending for a budget](https://aiplaybookgems.com/usecases/claude-categorize-a-month-of-spending-for-a-budget): Claude; Claude handles this as a draft: Use only the categories I provide. Put leftovers in Unsorted with the raw description. Do not move money between accounts. - [Cluster survey comments](https://aiplaybookgems.com/usecases/crewai-cluster-survey-comments): CrewAI; CrewAI handles this as a draft: Five themes and a quote for each. Do not identify people beyond the comment. - [Compare three flights I pasted](https://aiplaybookgems.com/usecases/chatgpt-agent-compare-three-flights-i-pasted): ChatGPT Agent; ChatGPT Agent handles this as a draft: Duration, stops, and bag fee if the text includes it. Blank if it does not. Do not buy a ticket. - [Compare two after-school programs from their PDFs](https://aiplaybookgems.com/usecases/claude-compare-two-after-school-programs-from-their-pdfs): Claude; Claude handles this as a draft: Price, hours, and pickup rules. Mark missing facts. Do not enroll. - [Compare two mattress return policies](https://aiplaybookgems.com/usecases/claude-compare-two-mattress-return-policies): Claude; Claude handles this as a draft: Trial length, who pays return shipping, and fees. Quotes only. Do not buy a mattress. - [Compare two papers on one question](https://aiplaybookgems.com/usecases/gemini-compare-two-papers-on-one-question): Gemini; Gemini handles this as a draft: Method and result, side by side, with the limit each paper states. Do not declare a winner without a quote. - [Draft a reply that asks for the missing log](https://aiplaybookgems.com/usecases/lindy-draft-a-reply-that-asks-for-the-missing-log): Lindy; Lindy handles this as a draft: Polite, specific about which log, no promised fix. Do not send it. - [Draft a short story outline from my premise](https://aiplaybookgems.com/usecases/gemini-draft-a-short-story-outline-from-my-premise): Gemini; Gemini handles this as a draft: Beats only. Keep my ending if I wrote one. Do not post it. - [Explain a function from my notes](https://aiplaybookgems.com/usecases/gemini-explain-a-function-from-my-notes): Gemini; Gemini handles this as a draft: Use the definition I wrote. Say if my note is incomplete. Do not browse for a different definition. - [Explain a medical EOB in plain language](https://aiplaybookgems.com/usecases/gemini-explain-a-medical-eob-in-plain-language): Gemini; Gemini handles this as a draft: Rewrite what I owe, what insurance paid, and which code is still pending. Do not guess a code you cannot see. Do not advise treatment. - [Extract action items from a customer call](https://aiplaybookgems.com/usecases/claude-cowork-extract-action-items-from-a-customer-call): Claude Cowork; Claude Cowork handles this as a draft: What we promised, by when. Separate their requests we did not accept. Do not email the customer. - [Extract feature requests without merging them](https://aiplaybookgems.com/usecases/claude-extract-feature-requests-without-merging-them): Claude; Claude handles this as a draft: Quote the request. Do not decide the roadmap. Do not tell the customer it is planned. - [Find subscriptions that renewed without a receipt](https://aiplaybookgems.com/usecases/hermes-agent-find-subscriptions-that-renewed-without-a-receipt): Hermes Agent; Hermes Agent handles this as a draft: From the card export I provide, list merchants that charged more than once, the cadence, and which ones have no matching receipt in the folder. Do not cancel any subscription. - [Have a Zapier Agent prepare CRM follow-ups and wait](https://aiplaybookgems.com/usecases/zapier-crm-review-list): Zapier Agents; Zapier Agents draft follow-ups for stale deals and park them for approval. The playbook does not let the agent send or edit opportunity amounts. - [Hold a week of meetings that lack an agenda](https://aiplaybookgems.com/usecases/hermes-agent-hold-a-week-of-meetings-that-lack-an-agenda): Hermes Agent; Hermes Agent handles this as a draft: Find events this week with an empty description. Draft a one-line question I could ask the organizer. Do not decline or edit the invite. - [Let an n8n agent sort a Slack channel into a digest](https://aiplaybookgems.com/usecases/n8n-slack-triage-agent): n8n; An n8n agent reads one Slack channel, classifies messages, and posts a digest only to you. It does not reply in the channel. - [List vaccination record gaps](https://aiplaybookgems.com/usecases/gemini-list-vaccination-record-gaps): Gemini; Gemini handles this as a draft: Compare the record I upload to the school's required list. Return only missing lines. Do not book a clinic. - [Make a medication list from labels I transcribe](https://aiplaybookgems.com/usecases/gemini-make-a-medication-list-from-labels-i-transcribe): Gemini; Gemini handles this as a draft: Name, dose, and time as printed. Flag conflicts I should ask about, without telling me to stop a medicine. I confirm with a clinician. - [Organize vaccination dates I typed](https://aiplaybookgems.com/usecases/perplexity-organize-vaccination-dates-i-typed): Perplexity; Perplexity handles this as a draft: A table. Mark missing dates. Do not book a pharmacy. - [Outline an API change for one field](https://aiplaybookgems.com/usecases/codex-cli-outline-an-api-change-for-one-field): Codex CLI; Codex CLI handles this as a draft: Old, new, and who reads it in the snippets I paste. Do not ship the API. - [Prep a budget variance note](https://aiplaybookgems.com/usecases/grok-bot-prep-a-budget-variance-note): Grok Bot; Grok Bot handles this as a draft: Lines that moved more than the percent I set, with the reason column if present. Do not change the budget. - [Prep a lab report outline](https://aiplaybookgems.com/usecases/hermes-agent-prep-a-lab-report-outline): Hermes Agent; Hermes Agent handles this as a draft: Sections the rubric requires and which of my notes fill them. Do not write the conclusion as fact. - [Prep questions for a high-ticket salesperson](https://aiplaybookgems.com/usecases/manus-prep-questions-for-a-high-ticket-salesperson): Manus; Manus handles this as a draft: Five questions from the spec sheet gaps. Do not contact the store. - [Prepare a security report from a user email](https://aiplaybookgems.com/usecases/langgraph-prepare-a-security-report-from-a-user-email): LangGraph; LangGraph handles this as a draft: What they saw, when, and what I should not click. Do not reply with a link. - [Propose a small patch for one failing test](https://aiplaybookgems.com/usecases/cursor-propose-a-small-patch-for-one-failing-test): Cursor; Cursor handles this as a draft: The smallest change that could explain the failure I paste. Do not commit. - [Refactor one module and land tests in the same session](https://aiplaybookgems.com/usecases/claude-code-refactor-with-tests): Claude Code; Claude Code reads a single module, writes characterization tests, then changes structure only where those tests pass. - [Score a CRM list and stop before anyone is emailed](https://aiplaybookgems.com/usecases/grok-bot-account-research): Grok Bot; Ask a Grok Bot to score accounts you paste in, name at most two contacts, and draft outreach that stays in a review file. Nothing is sent. - [Score a lead list against my ICP notes](https://aiplaybookgems.com/usecases/zapier-agents-score-a-lead-list-against-my-icp-notes): Zapier Agents; Zapier Agents handles this as a draft: Fit or not, with the ICP line you used. Unknown if the row is thin. Do not email leads. - [Sort a downloads folder of life admin PDFs](https://aiplaybookgems.com/usecases/claude-cowork-sort-a-downloads-folder-of-life-admin-pdfs): Claude Cowork; Claude Cowork handles this as a draft: Propose groups: tax, medical, housing, other. Do not delete. Leave deletion to me. - [Sort shared-household expenses for a settle-up](https://aiplaybookgems.com/usecases/gemini-sort-shared-household-expenses-for-a-settle-up): Gemini; Gemini handles this as a draft: From the shared sheet, who paid, and the split rule I write. Produce the net. Do not message my housemate. - [Triage school and activity emails](https://aiplaybookgems.com/usecases/grok-bot-triage-school-and-activity-emails): Grok Bot; Grok Bot handles this as a draft: Group by child and by deadline. Surface only items due in seven days. Do not reply to the school. - [Triage the inbox from Telegram without auto-sending](https://aiplaybookgems.com/usecases/openclaw-inbox-from-telegram): OpenClaw; OpenClaw runs on your machine and can meet you in Telegram. This playbook asks it to sort the morning mail and draft replies it is not allowed to send. - [Turn a broken-appliance note into a parts shortlist](https://aiplaybookgems.com/usecases/gemini-turn-a-broken-appliance-note-into-a-parts-shortlist): Gemini; Gemini handles this as a draft: Model number and the symptom. List parts that match that symptom on the pages I paste. Do not order parts. - [Turn a camp packet into a packing list and dates](https://aiplaybookgems.com/usecases/claude-cowork-turn-a-camp-packet-into-a-packing-list-and-dates): Claude Cowork; Claude Cowork handles this as a draft: Dates, forms, and what to pack. Quote the form deadline. Do not register. - [Turn a conference agenda into a personal schedule](https://aiplaybookgems.com/usecases/perplexity-turn-a-conference-agenda-into-a-personal-schedule): Perplexity; Perplexity handles this as a draft: Sessions I starred, with conflicts called out. Do not register. - [Turn a meeting transcript into decisions and owners](https://aiplaybookgems.com/usecases/zapier-agents-turn-a-meeting-transcript-into-decisions-and-owners): Zapier Agents; Zapier Agents handles this as a draft: Decision, owner, and date. Omit small talk. Do not assign work in the tracker. - [Turn a road-trip doc into daily stops](https://aiplaybookgems.com/usecases/gemini-turn-a-road-trip-doc-into-daily-stops): Gemini; Gemini handles this as a draft: Drive legs under the hour cap I set. Flag days that break the cap. Do not reserve lodging. - [Turn a ticket into a small Cursor patch](https://aiplaybookgems.com/usecases/cursor-ticket-to-patch): Cursor; Cursor's agent searches the repo, edits the files for one ticket, and runs the checks you name. The ticket's out-of-scope list is part of the prompt. - [Turn client feedback into a revision list](https://aiplaybookgems.com/usecases/manus-turn-client-feedback-into-a-revision-list): Manus; Manus handles this as a draft: Each note and the frame it refers to. Do not send the revision. - [Turn feedback on an essay into a fix list](https://aiplaybookgems.com/usecases/perplexity-turn-feedback-on-an-essay-into-a-fix-list): Perplexity; Perplexity handles this as a draft: Each comment and the sentence it points at. Do not submit the essay. - [Turn testimonials into usable one-liners](https://aiplaybookgems.com/usecases/perplexity-turn-testimonials-into-usable-one-liners): Perplexity; Perplexity handles this as a draft: Keep their words. Note if I lack permission to publish. Do not post. - [Turn user interviews into jobs-to-be-done lines](https://aiplaybookgems.com/usecases/manus-turn-user-interviews-into-jobs-to-be-done-lines): Manus; Manus handles this as a draft: When, I want to, so I can. Use their verbs. Do not roadmap it. - [Write a photo caption that states only what is visible](https://aiplaybookgems.com/usecases/claude-cowork-write-a-photo-caption-that-states-only-what-is-visible): Claude Cowork; Claude Cowork handles this as a draft: No story the pixels do not show. Do not post. - [Write a PR description from a diff summary](https://aiplaybookgems.com/usecases/github-copilot-write-a-pr-description-from-a-diff-summary): GitHub Copilot; GitHub Copilot handles this as a draft: What changed and how to test. Do not open the PR. - [A morning digest that only keeps items tied to written priorities](https://aiplaybookgems.com/usecases/grok-bot-morning-digest): Grok Bot; Grok Bot reads the channels you approve and returns a short list mapped to a priorities doc. It does not send messages or move meetings. - [A vendor memo from public pages, delivered as a file](https://aiplaybookgems.com/usecases/manus-vendor-memo): Manus; Manus is documented as an agent with its own computer. This playbook asks for a sourced vendor memo and a file, not a chatty opinion. - [Audit annual card fees against the fee schedule](https://aiplaybookgems.com/usecases/claude-cowork-audit-annual-card-fees-against-the-fee-schedule): Claude Cowork; Claude Cowork handles this as a draft: Match each fee on the statement to a line in the schedule I paste. List fees with no match. Do not dispute a charge. - [Build a gift shortlist under a cap](https://aiplaybookgems.com/usecases/chatgpt-agent-build-a-gift-shortlist-under-a-cap): ChatGPT Agent; ChatGPT Agent handles this as a draft: Three options under the cap from pages I provide, with the price on the page. Do not check out. - [Build a two-week study plan from a syllabus](https://aiplaybookgems.com/usecases/perplexity-build-a-two-week-study-plan-from-a-syllabus): Perplexity; Perplexity handles this as a draft: Dates and readings on the syllabus only. Do not email the instructor. - [Check a brokerage statement for fees](https://aiplaybookgems.com/usecases/grok-bot-check-a-brokerage-statement-for-fees): Grok Bot; Grok Bot handles this as a draft: List advisory fees and fund expense ratios printed on the statement. Skip tickers you cannot see. Do not trade. - [Check a README's commands against the repo](https://aiplaybookgems.com/usecases/claude-code-check-a-readme-s-commands-against-the-repo): Claude Code; Claude Code handles this as a draft: Which commands are missing a script entry. Do not rewrite the README until I agree. - [Cluster tickets that share a phrase](https://aiplaybookgems.com/usecases/langgraph-cluster-tickets-that-share-a-phrase): LangGraph; LangGraph handles this as a draft: Group by the error string I name. Count each group. Do not close tickets. - [Draft a handoff note for a teammate out tomorrow](https://aiplaybookgems.com/usecases/lindy-draft-a-handoff-note-for-a-teammate-out-tomorrow): Lindy; Lindy handles this as a draft: Open loops from the doc I paste and the one password process I describe without writing the password. Do not message them. - [Draft a noise complaint with the dates I logged](https://aiplaybookgems.com/usecases/openclaw-draft-a-noise-complaint-with-the-dates-i-logged): OpenClaw; OpenClaw handles this as a draft: Turn the log into a polite note with dates and times. No insults. Do not send it. - [Draft a pet-sitter sheet](https://aiplaybookgems.com/usecases/lindy-draft-a-pet-sitter-sheet): Lindy; Lindy handles this as a draft: Vet, food, and the emergency number I provide. One page. Do not text the sitter. - [Draft a podcast rundown](https://aiplaybookgems.com/usecases/claude-draft-a-podcast-rundown): Claude; Claude handles this as a draft: Segments and the question for each guest I named. Do not invite guests. - [Draft a portal message asking for a record](https://aiplaybookgems.com/usecases/claude-draft-a-portal-message-asking-for-a-record): Claude; Claude handles this as a draft: What record, what dates. Under 80 words. Do not send the portal message. - [Draft a price-match request](https://aiplaybookgems.com/usecases/manus-draft-a-price-match-request): Manus; Manus handles this as a draft: Quote the competitor price from the page I paste and the policy line from my store. Do not send it. - [Draft a refund ask from a receipt and a policy](https://aiplaybookgems.com/usecases/grok-bot-draft-a-refund-ask-from-a-receipt-and-a-policy): Grok Bot; Grok Bot handles this as a draft: Read the receipt and the return policy I paste. Draft the email I would send. Quote the policy line you rely on. Do not send the email. - [Draft a renewal risk note](https://aiplaybookgems.com/usecases/lindy-draft-a-renewal-risk-note): Lindy; Lindy handles this as a draft: Usage signals I paste and the open ask. Do not email the customer. - [Draft a speaker bio from a résumé](https://aiplaybookgems.com/usecases/crewai-draft-a-speaker-bio-from-a-r-sum): CrewAI; CrewAI handles this as a draft: 90 words, only jobs on the résumé. Do not publish the bio. - [Draft a test plan for a bug](https://aiplaybookgems.com/usecases/codex-cli-draft-a-test-plan-for-a-bug): Codex CLI; Codex CLI handles this as a draft: Cases, including the one that should fail today. Do not merge. - [Draft commit messages for a split](https://aiplaybookgems.com/usecases/devin-draft-commit-messages-for-a-split): Devin; Devin handles this as a draft: One message per hunk group I describe. Do not commit. - [Draft five outreach lines I can edit](https://aiplaybookgems.com/usecases/perplexity-draft-five-outreach-lines-i-can-edit): Perplexity; Perplexity handles this as a draft: Each line cites a fact from the page I provide. No sequence that sends itself. Do not send. - [Explain a lab PDF in plain words](https://aiplaybookgems.com/usecases/claude-cowork-explain-a-lab-pdf-in-plain-words): Claude Cowork; Claude Cowork handles this as a draft: Which values the report itself marks high or low. Do not interpret a normal range you cannot see. Not a diagnosis. - [Extract numbers from a report into a table](https://aiplaybookgems.com/usecases/manus-extract-numbers-from-a-report-into-a-table): Manus; Manus handles this as a draft: Figure, page, and the unit. Skip charts you cannot read. Do not update the live dashboard. - [Extract what a travel insurance policy actually covers](https://aiplaybookgems.com/usecases/manus-extract-what-a-travel-insurance-policy-actually-covers): Manus; Manus handles this as a draft: Delays, medical, and cancellations, each with the exclusion I should read. Do not buy the policy. - [Flag lifestyle creep against a baseline month](https://aiplaybookgems.com/usecases/claude-cowork-flag-lifestyle-creep-against-a-baseline-month): Claude Cowork; Claude Cowork handles this as a draft: Compare this month's categories to the baseline month I choose. Only comment on categories that moved more than 20 percent. Do not shame or editorialize. - [List debts with the rate I can see](https://aiplaybookgems.com/usecases/hermes-agent-list-debts-with-the-rate-i-can-see): Hermes Agent; Hermes Agent handles this as a draft: Make a table of balance, rate, and minimum from the statements I upload. Leave rate blank when the PDF hides it. Do not apply for a loan. - [Make a QA sheet for drafted replies](https://aiplaybookgems.com/usecases/zapier-agents-make-a-qa-sheet-for-drafted-replies): Zapier Agents; Zapier Agents handles this as a draft: Check tone, policy quote, and whether a promise slipped in. I send the reply. - [Make a RACI from a project brief](https://aiplaybookgems.com/usecases/hermes-agent-make-a-raci-from-a-project-brief): Hermes Agent; Hermes Agent handles this as a draft: Only the roles named in the brief. Do not add departments. Do not notify anyone. - [Plan a no-spend family weekend from what we have](https://aiplaybookgems.com/usecases/hermes-agent-plan-a-no-spend-family-weekend-from-what-we-have): Hermes Agent; Hermes Agent handles this as a draft: Activities from the constraints I give: weather, ages, and budget zero. Do not buy tickets. - [Point Codex CLI at one failing check](https://aiplaybookgems.com/usecases/codex-cli-failing-check): Codex CLI; Codex CLI investigates a single red check, patches it, and shows the command output. It does not rewrite the suite to go green. - [Prep a handover for weekend support](https://aiplaybookgems.com/usecases/n8n-prep-a-handover-for-weekend-support): n8n; n8n handles this as a draft: Open sev issues and the workaround already sent. Do not change severity. - [Prep a lightning talk outline](https://aiplaybookgems.com/usecases/cursor-prep-a-lightning-talk-outline): Cursor; Cursor handles this as a draft: Three beats and the example I already have. Do not record it. - [Prep a reading order for a new domain](https://aiplaybookgems.com/usecases/chatgpt-agent-prep-a-reading-order-for-a-new-domain): ChatGPT Agent; ChatGPT Agent handles this as a draft: Start here, then this, from the list I already collected. Do not add books I did not list. - [Prep a travel medical sheet](https://aiplaybookgems.com/usecases/openclaw-prep-a-travel-medical-sheet): OpenClaw; OpenClaw handles this as a draft: Allergies and the clinic note I paste. One page. Do not give medical advice. - [Propose two meeting times from chat and wait](https://aiplaybookgems.com/usecases/openclaw-calendar-hold): OpenClaw; From WhatsApp, Telegram, or another connected channel, OpenClaw reads the calendar and proposes holds. It does not send the invite until you confirm. - [Summarize a parent-teacher note](https://aiplaybookgems.com/usecases/gemini-summarize-a-parent-teacher-note): Gemini; Gemini handles this as a draft: What the teacher asked, by when, and what is just context. Do not reply. - [Timeline a company from filings I paste](https://aiplaybookgems.com/usecases/perplexity-timeline-a-company-from-filings-i-paste): Perplexity; Perplexity handles this as a draft: Date and event only. Do not infer strategy. - [Triage a support queue into a list Lindy does not send](https://aiplaybookgems.com/usecases/lindy-support-triage): Lindy; Lindy's site lists support-queue triage as a use case. This playbook produces a prioritized list and draft replies that wait for a person. - [Turn a disruption text into next actions](https://aiplaybookgems.com/usecases/gemini-turn-a-disruption-text-into-next-actions): Gemini; Gemini handles this as a draft: What changed, the deadline to act, and the two options the airline actually offered. Do not accept a voucher. - [Turn a group chat export into decisions only](https://aiplaybookgems.com/usecases/openclaw-turn-a-group-chat-export-into-decisions-only): OpenClaw; OpenClaw handles this as a draft: List decisions and open questions. Drop banter. Do not post back to the chat. - [Turn a notes app dump into a weekend list](https://aiplaybookgems.com/usecases/claude-cowork-turn-a-notes-app-dump-into-a-weekend-list): Claude Cowork; Claude Cowork handles this as a draft: Extract errands, each with a place if I wrote one. Drop anything that is a wish, not a task. Do not message anyone. - [Turn a recipe into a shopping list with substitutions](https://aiplaybookgems.com/usecases/perplexity-turn-a-recipe-into-a-shopping-list-with-substitutions): Perplexity; Perplexity handles this as a draft: Flag ingredients I said I will not buy and propose one substitute from my pantry list. Do not order groceries. - [Turn a wearable export into a one-week note](https://aiplaybookgems.com/usecases/hermes-agent-turn-a-wearable-export-into-a-one-week-note): Hermes Agent; Hermes Agent handles this as a draft: Sleep and steps totals that are in the file. Skip metrics the file lacks. Not medical advice. - [Turn a webinar transcript into a recap post](https://aiplaybookgems.com/usecases/grok-bot-turn-a-webinar-transcript-into-a-recap-post): Grok Bot; Grok Bot handles this as a draft: Three points and a question. No attendance numbers I did not give. Do not post. - [Turn bank alerts into a weekly money memo](https://aiplaybookgems.com/usecases/n8n-turn-bank-alerts-into-a-weekly-money-memo): n8n; n8n handles this as a draft: Cluster alerts by merchant. Lead with anything over the threshold I set. Ignore marketing mail. Do not move money. - [Turn office-hour notes into a concept map](https://aiplaybookgems.com/usecases/claude-turn-office-hour-notes-into-a-concept-map): Claude; Claude handles this as a draft: Only terms I wrote down. Do not add chapters we have not reached. - [Write a close-the-loop note after an incident](https://aiplaybookgems.com/usecases/n8n-write-a-close-the-loop-note-after-an-incident): n8n; n8n handles this as a draft: Timeline and the customer impact sentence, from the doc. Do not email customers. - [Write alt lines for a headline I dislike](https://aiplaybookgems.com/usecases/manus-write-alt-lines-for-a-headline-i-dislike): Manus; Manus handles this as a draft: Five options in the voice notes I gave. Do not update the site. - [Audit a cart for duplicate subscriptions](https://aiplaybookgems.com/usecases/claude-audit-a-cart-for-duplicate-subscriptions): Claude; Claude handles this as a draft: Point at items that overlap with subscriptions I listed. Do not purchase the cart. - [Break a video idea into shots](https://aiplaybookgems.com/usecases/crewai-break-a-video-idea-into-shots): CrewAI; CrewAI handles this as a draft: Shot, what is on screen, and the line. No stock I did not name. Do not publish the video. - [Build a medication schedule for a visit](https://aiplaybookgems.com/usecases/claude-cowork-build-a-medication-schedule-for-a-visit): Claude Cowork; Claude Cowork handles this as a draft: Times and doses from the label text I paste. Ask me if two labels disagree. Do not change a dose. - [Build a packing list for a week at home repairs](https://aiplaybookgems.com/usecases/hermes-agent-build-a-packing-list-for-a-week-at-home-repairs): Hermes Agent; Hermes Agent handles this as a draft: Tools and materials named in the estimate I paste. Flag vague lines like miscellaneous. Do not hire anyone. - [Build a school-supply list against a store aisle](https://aiplaybookgems.com/usecases/perplexity-build-a-school-supply-list-against-a-store-aisle): Perplexity; Perplexity handles this as a draft: Map the teacher's list to item names. Mark anything ambiguous like folder. Do not check out. - [Build a symptom timeline from my notes](https://aiplaybookgems.com/usecases/hermes-agent-build-a-symptom-timeline-from-my-notes): Hermes Agent; Hermes Agent handles this as a draft: Date and symptom in my words. Do not suggest a disease. - [Check a bibliography for missing fields](https://aiplaybookgems.com/usecases/hermes-agent-check-a-bibliography-for-missing-fields): Hermes Agent; Hermes Agent handles this as a draft: Which entries lack a year or a title. Do not fetch new papers. - [Check a draft for unsourced sentences](https://aiplaybookgems.com/usecases/gemini-check-a-draft-for-unsourced-sentences): Gemini; Gemini handles this as a draft: List sentences that state a number or a cause. Do not add sources you did not open. - [Check accessibility of a component from the markup](https://aiplaybookgems.com/usecases/github-copilot-check-accessibility-of-a-component-from-the-markup): GitHub Copilot; GitHub Copilot handles this as a draft: Name, role, and keyboard notes from the snippet. Do not restyle the app. - [Compare two neighborhoods for a short stay](https://aiplaybookgems.com/usecases/chatgpt-agent-compare-two-neighborhoods-for-a-short-stay): ChatGPT Agent; ChatGPT Agent handles this as a draft: From the two guides I paste: walk to the station, noise, and groceries. Do not book a stay. - [Draft a paycheck allocation I can edit](https://aiplaybookgems.com/usecases/grok-bot-draft-a-paycheck-allocation-i-can-edit): Grok Bot; Grok Bot handles this as a draft: Split the net pay I give you across the buckets I name. Show the remainder. Do not change direct deposit. - [Draft a retro from ticket comments](https://aiplaybookgems.com/usecases/grok-bot-draft-a-retro-from-ticket-comments): Grok Bot; Grok Bot handles this as a draft: What repeated, with links or ids I included. No blame lines. Do not post the retro. - [Draft a status page sentence](https://aiplaybookgems.com/usecases/lindy-draft-a-status-page-sentence): Lindy; Lindy handles this as a draft: Impact and the next update time I choose. No cause if we do not know it. Do not post the status. - [Explain a COBRA offer before the deadline](https://aiplaybookgems.com/usecases/gemini-explain-a-cobra-offer-before-the-deadline): Gemini; Gemini handles this as a draft: State the monthly cost, who is covered, and the date the offer expires, using only the letter. Do not elect coverage. - [Explain a prior-authorization letter](https://aiplaybookgems.com/usecases/gemini-explain-a-prior-authorization-letter): Gemini; Gemini handles this as a draft: What was denied, the appeal deadline, and the address. Do not appeal for me. - [List citations I still need for a draft](https://aiplaybookgems.com/usecases/gemini-list-citations-i-still-need-for-a-draft): Gemini; Gemini handles this as a draft: Claims in my draft with no footnote. Do not invent sources. - [List contradictions in a set of articles](https://aiplaybookgems.com/usecases/chatgpt-agent-list-contradictions-in-a-set-of-articles): ChatGPT Agent; ChatGPT Agent handles this as a draft: Claim A, claim B, and the two titles. Do not smooth them into one story. - [Make a bill calendar from due dates](https://aiplaybookgems.com/usecases/claude-make-a-bill-calendar-from-due-dates): Claude; Claude handles this as a draft: List the bill, amount, and due day. Mark anything without a due date. Do not enroll in autopay. - [Make a content calendar week from topics I listed](https://aiplaybookgems.com/usecases/manus-make-a-content-calendar-week-from-topics-i-listed): Manus; Manus handles this as a draft: Title and the source I must cite. No invented news. Do not schedule posts. - [Make a packing list for a climate I name](https://aiplaybookgems.com/usecases/manus-make-a-packing-list-for-a-climate-i-name): Manus; Manus handles this as a draft: Clothes and documents. Use the weather notes I paste, not a live lookup unless I included one. Do not buy gear. - [Make a shared birthday list with budget caps](https://aiplaybookgems.com/usecases/hermes-agent-make-a-shared-birthday-list-with-budget-caps): Hermes Agent; Hermes Agent handles this as a draft: Person, date, idea, and cap. Skip anyone without a date. Do not buy gifts. - [Make flashcards from a lecture outline](https://aiplaybookgems.com/usecases/cursor-make-flashcards-from-a-lecture-outline): Cursor; Cursor handles this as a draft: Term and the definition in the outline. Skip terms without a definition. Do not invent facts. - [Map a pricing page to my questions](https://aiplaybookgems.com/usecases/zapier-agents-map-a-pricing-page-to-my-questions): Zapier Agents; Zapier Agents handles this as a draft: What is public and what is contact-sales. Do not fill their form. - [Note the limits in a benchmark writeup](https://aiplaybookgems.com/usecases/crewai-note-the-limits-in-a-benchmark-writeup): CrewAI; CrewAI handles this as a draft: Hardware, date, and what the authors say they did not test. Do not treat it as a purchase recommendation. - [Outline a case study from notes](https://aiplaybookgems.com/usecases/claude-cowork-outline-a-case-study-from-notes): Claude Cowork; Claude Cowork handles this as a draft: Problem, what we did, and the metric I supplied. No extra metrics. Do not get customer approval for me. - [Prep a home-maintenance log from photos](https://aiplaybookgems.com/usecases/lindy-prep-a-home-maintenance-log-from-photos): Lindy; Lindy handles this as a draft: For each photo I attach, write the fixture, the visible issue, and whether it looks urgent. Say when a photo is unclear. Do not book a contractor. - [Prep a house sitter note before a trip](https://aiplaybookgems.com/usecases/perplexity-prep-a-house-sitter-note-before-a-trip): Perplexity; Perplexity handles this as a draft: Plants, bins, and the neighbor's number I provide. Do not text the sitter. - [Prep a second-opinion packet list](https://aiplaybookgems.com/usecases/perplexity-prep-a-second-opinion-packet-list): Perplexity; Perplexity handles this as a draft: Which records the specialist's page says to bring. Do not request records. - [Prepare questions for a benefits enrollment call](https://aiplaybookgems.com/usecases/hermes-agent-prepare-questions-for-a-benefits-enrollment-call): Hermes Agent; Hermes Agent handles this as a draft: From the plan summary, list deductibles, out-of-network rules, and three questions I should ask before I enroll. Do not enroll me. - [Propose types for an untyped function](https://aiplaybookgems.com/usecases/cursor-propose-types-for-an-untyped-function): Cursor; Cursor handles this as a draft: From the tests I paste. Do not change runtime behavior. - [Review a diff for risky calls](https://aiplaybookgems.com/usecases/devin-review-a-diff-for-risky-calls): Devin; Devin handles this as a draft: Network, auth, and migrations. Quote the lines. Do not approve the PR. - [Sort an ops inbox by SLA risk](https://aiplaybookgems.com/usecases/claude-cowork-sort-an-ops-inbox-by-sla-risk): Claude Cowork; Claude Cowork handles this as a draft: Which threads mention a deadline inside 24 hours. Do not reply. - [Sort hiring-loop feedback into themes](https://aiplaybookgems.com/usecases/zapier-agents-sort-hiring-loop-feedback-into-themes): Zapier Agents; Zapier Agents handles this as a draft: Strengths and risks mentioned by more than one interviewer. Do not advance a candidate. - [Summarize a custody calendar change](https://aiplaybookgems.com/usecases/claude-summarize-a-custody-calendar-change): Claude; Claude handles this as a draft: What moved, which week, and the sentence I should confirm with the other parent. Do not send that sentence. - [Summarize a partner escalation](https://aiplaybookgems.com/usecases/claude-summarize-a-partner-escalation): Claude; Claude handles this as a draft: Who is blocked and the last timestamp. Do not escalate further. - [Tighten a landing paragraph](https://aiplaybookgems.com/usecases/gemini-tighten-a-landing-paragraph): Gemini; Gemini handles this as a draft: Shorter, same facts. Show me the cuts. Do not ship the page. - [Turn a call transcript into a CRM note](https://aiplaybookgems.com/usecases/lindy-turn-a-call-transcript-into-a-crm-note): Lindy; Lindy handles this as a draft: Pain, next step, and date. No fields I did not hear. Do not update the CRM. - [Turn a help-center article into a shorter macro](https://aiplaybookgems.com/usecases/zapier-agents-turn-a-help-center-article-into-a-shorter-macro): Zapier Agents; Zapier Agents handles this as a draft: The steps only, with the article title at the top. Do not publish the macro. - [Turn a project materials list into a store run](https://aiplaybookgems.com/usecases/gemini-turn-a-project-materials-list-into-a-store-run): Gemini; Gemini handles this as a draft: Group by store if I named one. Quantities from my list only. Do not reserve inventory. - [Turn a year of utility bills into a rate check](https://aiplaybookgems.com/usecases/n8n-turn-a-year-of-utility-bills-into-a-rate-check): n8n; n8n handles this as a draft: Summarize usage and total paid by month. Point at the month the rate jumped and the line on the bill that explains it. Do not switch providers or submit a form. - [Weekly home reset from a standing checklist](https://aiplaybookgems.com/usecases/grok-bot-weekly-home-reset-from-a-standing-checklist): Grok Bot; Grok Bot handles this as a draft: Compare the checklist to the notes I add this week. Return only the undone items. Do not nag other people in the house. - [Write the rules for a new sinking fund](https://aiplaybookgems.com/usecases/gemini-write-the-rules-for-a-new-sinking-fund): Gemini; Gemini handles this as a draft: Name the goal, monthly amount, and the date it would fill, using the numbers I provide. Do not open the account. - [Draft a pull request summary from the diff](https://aiplaybookgems.com/usecases/copilot-pr-summary): GitHub Copilot; GitHub Copilot writes a PR summary a reviewer can check against the diff: what changed, how to test, and risks. It does not approve the PR. - [Reconcile a week of expenses without touching payouts](https://aiplaybookgems.com/usecases/grok-bot-expense-week): Grok Bot; The bot builds a weekly expense summary, matches receipts it can see, and drafts one follow-up per owner. It does not send or reimburse. - [Research a public page set and keep a source log](https://aiplaybookgems.com/usecases/chatgpt-agent-public-research): ChatGPT Agent; A computer-use style agent visits a list of public URLs and returns notes with the page each claim came from. It does not log in. - [Upgrade one dependency and read the changelog first](https://aiplaybookgems.com/usecases/claude-code-dependency-upgrade): Claude Code; Claude Code summarizes breaking changes for a single dependency, applies the upgrade, and runs the existing tests. Major bumps stay behind a question. - [Build a glossary from a technical packet](https://aiplaybookgems.com/usecases/crewai-build-a-glossary-from-a-technical-packet): CrewAI; CrewAI handles this as a draft: Term and the sentence in the packet. Do not add outside definitions. - [Build a vendor contact sheet from contracts](https://aiplaybookgems.com/usecases/hermes-agent-build-a-vendor-contact-sheet-from-contracts): Hermes Agent; Hermes Agent handles this as a draft: Name, renewal, and notice period. Blank if the PDF lacks it. Do not email vendors. - [Check a coupon against the exclusions](https://aiplaybookgems.com/usecases/manus-check-a-coupon-against-the-exclusions): Manus; Manus handles this as a draft: Say whether the item I named is excluded, quoting the exclusion. Do not apply the coupon. - [Compare three vendor security pages](https://aiplaybookgems.com/usecases/manus-compare-three-vendor-security-pages): Manus; Manus handles this as a draft: Certifications named on the page. Blank if absent. Do not fill a security form. - [Compare two clinic instructions I was given](https://aiplaybookgems.com/usecases/perplexity-compare-two-clinic-instructions-i-was-given): Perplexity; Perplexity handles this as a draft: Where they disagree. Quote both. Do not pick a treatment. - [Compare two competitor homepages](https://aiplaybookgems.com/usecases/manus-compare-two-competitor-homepages): Manus; Manus handles this as a draft: Offer, proof, and call to action, quoted. No traffic claims. Do not publish a teardown. - [Compare two explanations of the same idea](https://aiplaybookgems.com/usecases/hermes-agent-compare-two-explanations-of-the-same-idea): Hermes Agent; Hermes Agent handles this as a draft: Where they agree and the one sentence they do not. Do not pick a side without a quote. - [Compare two SOPs for contradictions](https://aiplaybookgems.com/usecases/n8n-compare-two-sops-for-contradictions): n8n; n8n handles this as a draft: Side-by-side disagreements. Ignore matching lines. Do not delete an SOP. - [Draft a building-management note about a leak](https://aiplaybookgems.com/usecases/grok-bot-draft-a-building-management-note-about-a-leak): Grok Bot; Grok Bot handles this as a draft: Use my notes and the photo description. Ask for a time window. Keep it under 120 words. Do not send it. - [Draft a caregiver handoff](https://aiplaybookgems.com/usecases/claude-draft-a-caregiver-handoff): Claude; Claude handles this as a draft: Meds, routines, and the number to call, using only my notes. Do not message the caregiver. - [Draft a delay reimbursement note](https://aiplaybookgems.com/usecases/gemini-draft-a-delay-reimbursement-note): Gemini; Gemini handles this as a draft: Timeline from my notes and the rule I paste. Attach list of receipts I name. Do not file the claim. - [Draft a partner intro I will send myself](https://aiplaybookgems.com/usecases/perplexity-draft-a-partner-intro-i-will-send-myself): Perplexity; Perplexity handles this as a draft: Why them, why us, one ask. Under 120 words. Do not send. - [Edit a transcript into a tight script](https://aiplaybookgems.com/usecases/claude-cowork-edit-a-transcript-into-a-tight-script): Claude Cowork; Claude Cowork handles this as a draft: Cut repeats. Do not add claims. Do not upload it. - [Explain a compiler error next to the line](https://aiplaybookgems.com/usecases/perplexity-explain-a-compiler-error-next-to-the-line): Perplexity; Perplexity handles this as a draft: What the message says and the line I paste. Do not rewrite the whole project. - [Explain a credit-card promo before it expires](https://aiplaybookgems.com/usecases/hermes-agent-explain-a-credit-card-promo-before-it-expires): Hermes Agent; Hermes Agent handles this as a draft: State the promo APR, the expiration, and what the letter says happens next. Quote the sentence. Do not apply for a card. - [Explain a module to a new teammate](https://aiplaybookgems.com/usecases/github-copilot-explain-a-module-to-a-new-teammate): GitHub Copilot; GitHub Copilot handles this as a draft: What it owns and what it must not import, from the code. Do not refactor it. - [Extract renewal dates from insurance mail](https://aiplaybookgems.com/usecases/claude-cowork-extract-renewal-dates-from-insurance-mail): Claude Cowork; Claude Cowork handles this as a draft: Make a table of policy, carrier, and renewal date. Skip mail that is an advertisement. Do not change a policy. - [Extract reservation numbers from a mail folder](https://aiplaybookgems.com/usecases/perplexity-extract-reservation-numbers-from-a-mail-folder): Perplexity; Perplexity handles this as a draft: Hotel, flight, and confirmation code. Skip marketing. Do not modify a booking. - [Find annual charges hiding in a year of CSV rows](https://aiplaybookgems.com/usecases/gemini-find-annual-charges-hiding-in-a-year-of-csv-rows): Gemini; Gemini handles this as a draft: List merchants billed about once a year and the month they hit. Do not cancel them. - [Find tickets waiting on us more than two days](https://aiplaybookgems.com/usecases/claude-find-tickets-waiting-on-us-more-than-two-days): Claude; Claude handles this as a draft: List id, age, and the last agent line. Do not ping the customer. - [List error paths in a handler](https://aiplaybookgems.com/usecases/codex-cli-list-error-paths-in-a-handler): Codex CLI; Codex CLI handles this as a draft: What the user sees for each branch. Do not deploy. - [Make a food log summary for a dietitian](https://aiplaybookgems.com/usecases/claude-cowork-make-a-food-log-summary-for-a-dietitian): Claude Cowork; Claude Cowork handles this as a draft: Foods and times I wrote. No advice. Do not email the dietitian. - [Prep a family emergency card](https://aiplaybookgems.com/usecases/gemini-prep-a-family-emergency-card): Gemini; Gemini handles this as a draft: Contacts and the meeting place I name. One card, large type. Do not publish it. - [Prep a Schedule C packet from a bank CSV](https://aiplaybookgems.com/usecases/gemini-prep-a-schedule-c-packet-from-a-bank-csv): Gemini; Gemini handles this as a draft: Group expenses into categories I name, mark rows you are unsure about, and total only the rows with a category. Do not file a return or connect to tax software. - [Reconcile a reimbursement report to receipts](https://aiplaybookgems.com/usecases/grok-bot-reconcile-a-reimbursement-report-to-receipts): Grok Bot; Grok Bot handles this as a draft: Each report row needs a receipt filename or the word missing. Totals must use only matched rows. Do not submit the report. - [Sort billing tickets that need finance](https://aiplaybookgems.com/usecases/langgraph-sort-billing-tickets-that-need-finance): LangGraph; LangGraph handles this as a draft: Which ones mention a charge id. Do not refund. - [Sort wish-list items by actual need](https://aiplaybookgems.com/usecases/claude-sort-wish-list-items-by-actual-need): Claude; Claude handles this as a draft: Label each item need, nice, or duplicate of something I said I own. Do not purchase. - [Turn a brand voice note into a checklist](https://aiplaybookgems.com/usecases/manus-turn-a-brand-voice-note-into-a-checklist): Manus; Manus handles this as a draft: Do and don't from my note only. Do not rewrite the guidelines I did not write. - [A sourced market scan with the holes left visible](https://aiplaybookgems.com/usecases/perplexity-market-scan): Perplexity; Perplexity searches current public sources and returns a scan that separates numbers it found from numbers it did not. - [Ask Devin for a pull request from a tightly written issue](https://aiplaybookgems.com/usecases/devin-issue-to-pr): Devin; Devin picks up one issue and opens a PR. The playbook tells it which repo, which commands, and what not to touch. - [Hand an X thread to a bot and get the brief in the app](https://aiplaybookgems.com/usecases/grok-bot-tag-on-x): Grok Bot; Tag @bot on a post so Grok Bot picks up the thread. The public reply is only a confirmation. The brief stays in the bot conversation. - [Write the regression test for a bug you can reproduce](https://aiplaybookgems.com/usecases/cursor-regression-tests): Cursor; Cursor Agent adds a failing test for a known bug, then the fix, and stops if the test never failed. - [Build an on-call handoff](https://aiplaybookgems.com/usecases/zapier-agents-build-an-on-call-handoff): Zapier Agents; Zapier Agents handles this as a draft: Open incidents and the last action, from the notes I paste. Do not page anyone. - [Cluster campaign results I exported](https://aiplaybookgems.com/usecases/lindy-cluster-campaign-results-i-exported): Lindy; Lindy handles this as a draft: Which message had the replies, using the columns I name. Do not launch a new campaign. - [Compare two insurance EOBs for the same visit](https://aiplaybookgems.com/usecases/claude-cowork-compare-two-insurance-eobs-for-the-same-visit): Claude Cowork; Claude Cowork handles this as a draft: Explain what each document says was billed, allowed, and owed. List contradictions instead of averaging them. Do not call the insurer. - [Compare used versus new from listings I paste](https://aiplaybookgems.com/usecases/chatgpt-agent-compare-used-versus-new-from-listings-i-paste): ChatGPT Agent; ChatGPT Agent handles this as a draft: Price, seller type, and return rule. Skip listings without a price. Do not message sellers. - [Draft a charity-donation log for taxes](https://aiplaybookgems.com/usecases/n8n-draft-a-charity-donation-log-for-taxes): n8n; n8n handles this as a draft: From the confirmation emails I paste, make a table of date, organization, and amount. Skip anything without an amount. Do not file taxes. - [Draft a hotel request for an early check-in](https://aiplaybookgems.com/usecases/openclaw-draft-a-hotel-request-for-an-early-check-in): OpenClaw; OpenClaw handles this as a draft: One short note. Mention the arrival time I give. Do not send it. - [Draft a launch email from a changelog](https://aiplaybookgems.com/usecases/grok-bot-draft-a-launch-email-from-a-changelog): Grok Bot; Grok Bot handles this as a draft: Three bullets from the changelog only. Do not send the email. - [Draft a rollback note](https://aiplaybookgems.com/usecases/devin-draft-a-rollback-note): Devin; Devin handles this as a draft: The command in the runbook I paste and the signal that means stop. Do not execute it. - [Draft a spending pause for one category](https://aiplaybookgems.com/usecases/grok-bot-draft-a-spending-pause-for-one-category): Grok Bot; Grok Bot handles this as a draft: Propose a 30-day rule for the category I name, using last month's actual total. Do not lock a card. - [Extract form fields from a permission slip](https://aiplaybookgems.com/usecases/openclaw-extract-form-fields-from-a-permission-slip): OpenClaw; OpenClaw handles this as a draft: List every blank and the fact I still need to supply. Do not sign. - [List upgrade steps from a changelog](https://aiplaybookgems.com/usecases/claude-code-list-upgrade-steps-from-a-changelog): Claude Code; Claude Code handles this as a draft: Breaking changes that mention the version I am on. Do not run the upgrade. - [Make a group dinner shortlist](https://aiplaybookgems.com/usecases/manus-make-a-group-dinner-shortlist): Manus; Manus handles this as a draft: Three places from the links I provide, with the reservation rule on the page. Do not book a table. - [Make a moodboard checklist](https://aiplaybookgems.com/usecases/claude-make-a-moodboard-checklist): Claude; Claude handles this as a draft: Colors and references I listed, grouped. Do not download images. - [Make a reading schedule around my shifts](https://aiplaybookgems.com/usecases/cursor-make-a-reading-schedule-around-my-shifts): Cursor; Cursor handles this as a draft: Use the hours I said I have. Do not move my shifts. - [Plan a grocery list from a week's dinners](https://aiplaybookgems.com/usecases/openclaw-plan-a-grocery-list-from-a-week-s-dinners): OpenClaw; OpenClaw handles this as a draft: Deduplicate ingredients from the meals I name. Mark staples I said I already have. Do not order delivery. - [Prep a pre-op checklist from the hospital packet](https://aiplaybookgems.com/usecases/claude-prep-a-pre-op-checklist-from-the-hospital-packet): Claude; Claude handles this as a draft: Stop-food time, what to bring, and the phone number on the packet. Do not call the hospital. - [Propose a weekly ops checklist from last month's misses](https://aiplaybookgems.com/usecases/grok-bot-propose-a-weekly-ops-checklist-from-last-month-s-misses): Grok Bot; Grok Bot handles this as a draft: Use the incident notes I paste. Five checks, each tied to a miss. Do not change production. - [Pull exercise notes into a weekly view](https://aiplaybookgems.com/usecases/hermes-agent-pull-exercise-notes-into-a-weekly-view): Hermes Agent; Hermes Agent handles this as a draft: What I logged. No new plan. Do not invent workouts. - [Sort allowance and chore notes](https://aiplaybookgems.com/usecases/hermes-agent-sort-allowance-and-chore-notes): Hermes Agent; Hermes Agent handles this as a draft: Child, chore, and the amount I already decided. Do not invent amounts. Do not transfer money. - [Storyboard a 30-second demo](https://aiplaybookgems.com/usecases/crewai-storyboard-a-30-second-demo): CrewAI; CrewAI handles this as a draft: Six frames and the UI I said is real. Do not record the demo. - [Summarize a angry thread for a manager](https://aiplaybookgems.com/usecases/n8n-summarize-a-angry-thread-for-a-manager): n8n; n8n handles this as a draft: What they asked, what we did, and the open promise. Do not reply in the thread. - [Summarize reviews for one failure mode](https://aiplaybookgems.com/usecases/perplexity-summarize-reviews-for-one-failure-mode): Perplexity; Perplexity handles this as a draft: From the reviews I paste, only comments about the failure I name. Count them. Do not scrape a site I did not paste. - [Turn a call recording note into a case comment](https://aiplaybookgems.com/usecases/zapier-agents-turn-a-call-recording-note-into-a-case-comment): Zapier Agents; Zapier Agents handles this as a draft: Facts only, under 150 words. Do not email the customer. - [Turn a dataset dictionary into questions](https://aiplaybookgems.com/usecases/perplexity-turn-a-dataset-dictionary-into-questions): Perplexity; Perplexity handles this as a draft: Which fields are missing descriptions. Do not query production data. - [Turn a literature table into gaps](https://aiplaybookgems.com/usecases/chatgpt-agent-turn-a-literature-table-into-gaps): ChatGPT Agent; ChatGPT Agent handles this as a draft: What the rows do not cover, in one list. Do not claim the gap is unstudied beyond this table. - [Turn a problem set into a hint ladder](https://aiplaybookgems.com/usecases/claude-turn-a-problem-set-into-a-hint-ladder): Claude; Claude handles this as a draft: Hints that do not reveal the final answer. I still solve it. - [Turn voice memos into a household decision list](https://aiplaybookgems.com/usecases/lindy-turn-voice-memos-into-a-household-decision-list): Lindy; Lindy handles this as a draft: Transcribe the decisions, the options, and who I said should weigh in. Do not message those people. - [Write a one-page net-worth snapshot](https://aiplaybookgems.com/usecases/claude-cowork-write-a-one-page-net-worth-snapshot): Claude Cowork; Claude Cowork handles this as a draft: Assets and debts I paste, with a note when a number is an estimate I marked as such. Do not pull accounts on your own. - [A two-role CrewAI crew that argues with its own draft](https://aiplaybookgems.com/usecases/crewai-research-crew): CrewAI; A researcher agent gathers sources and a reviewer agent rejects unsourced lines. You run it locally from the CrewAI docs' model. - [Answer questions from a Google Sheet without editing it](https://aiplaybookgems.com/usecases/n8n-sheets-answers): n8n; An n8n agent uses a sheet as its knowledge source and answers in a private chat. The sheet stays read-only. - [Remind me to check in, then stop for a yes](https://aiplaybookgems.com/usecases/openclaw-flight-checkin): OpenClaw; OpenClaw's site says it can check you in for flights from chat. This playbook keeps the reminder, and requires a yes before any airline page is submitted. - [Ask Copilot for a test plan, then run it yourself](https://aiplaybookgems.com/usecases/copilot-test-plan): GitHub Copilot; Copilot lists cases for a change you describe. You run them. The plan is not permission to skip the ones that are slow. - [Collect a competitor pack into one folder of files](https://aiplaybookgems.com/usecases/manus-competitor-pack): Manus; Manus gathers public pages, pricing screenshots, and a contradiction list into files you can drop into a doc. It does not publish anything. - [Map a repo for someone who cloned it this morning](https://aiplaybookgems.com/usecases/claude-code-repo-map): Claude Code; Claude Code writes a one-page map of how to run, test, and change the system. It does not edit application code. - [Plan the week from the calendar without moving it](https://aiplaybookgems.com/usecases/lindy-plan-the-week): Lindy; Lindy's site lists 'Plan your week'. This playbook returns a weekly plan and proposed moves. The calendar stays put until you edit it. - [Build a carry-on kit for a red-eye with a child](https://aiplaybookgems.com/usecases/perplexity-build-a-carry-on-kit-for-a-red-eye-with-a-child): Perplexity; Perplexity handles this as a draft: Items from my house list that fit a personal item. Do not shop. - [Compare two savings yields I paste](https://aiplaybookgems.com/usecases/claude-compare-two-savings-yields-i-paste): Claude; Claude handles this as a draft: Table the rate, the cap, and the fee. Say which number you could not find on the page. Do not open an account. - [Draft a bug report from a customer description](https://aiplaybookgems.com/usecases/lindy-draft-a-bug-report-from-a-customer-description): Lindy; Lindy handles this as a draft: Steps, expected, actual. Mark unknowns. Do not file it if the tracker needs an engineer. - [Draft a discussion post from my notes](https://aiplaybookgems.com/usecases/hermes-agent-draft-a-discussion-post-from-my-notes): Hermes Agent; Hermes Agent handles this as a draft: My point and one quote from the reading. Do not post it. - [Draft a marketplace listing for something I own](https://aiplaybookgems.com/usecases/claude-draft-a-marketplace-listing-for-something-i-own): Claude; Claude handles this as a draft: Title, condition from my words, and flaws I listed. No invented features. Do not post the listing. - [Draft a newsletter intro from three links](https://aiplaybookgems.com/usecases/gemini-draft-a-newsletter-intro-from-three-links): Gemini; Gemini handles this as a draft: One sentence per link using the title I pasted. Do not send the newsletter. - [Draft a teacher absence email](https://aiplaybookgems.com/usecases/claude-draft-a-teacher-absence-email): Claude; Claude handles this as a draft: Dates and the work I want sent home, in my voice, under 100 words. Do not send it. - [Draft show notes from a timestamp list](https://aiplaybookgems.com/usecases/claude-cowork-draft-show-notes-from-a-timestamp-list): Claude Cowork; Claude Cowork handles this as a draft: Topic at each timestamp I marked. Do not publish. - [Extract definitions of a metric across docs](https://aiplaybookgems.com/usecases/crewai-extract-definitions-of-a-metric-across-docs): CrewAI; CrewAI handles this as a draft: Each doc's definition, quoted. Do not pick one as official. - [Extract sizes from a clothing page for a gift](https://aiplaybookgems.com/usecases/gemini-extract-sizes-from-a-clothing-page-for-a-gift): Gemini; Gemini handles this as a draft: The size chart numbers and the one question I should ask the recipient. Do not order. - [List insurance questions before a procedure](https://aiplaybookgems.com/usecases/gemini-list-insurance-questions-before-a-procedure): Gemini; Gemini handles this as a draft: From the estimate: what is covered and what the estimate says I might owe. Do not schedule the procedure. - [List macros that contradict the new policy](https://aiplaybookgems.com/usecases/claude-list-macros-that-contradict-the-new-policy): Claude; Claude handles this as a draft: Quote both lines. Do not edit the macros. - [List objections from lost-deal notes](https://aiplaybookgems.com/usecases/zapier-agents-list-objections-from-lost-deal-notes): Zapier Agents; Zapier Agents handles this as a draft: The objection in their words and how often it appears. Do not write a battlecard yet. - [List unclaimed-property search results worth opening](https://aiplaybookgems.com/usecases/hermes-agent-list-unclaimed-property-search-results-worth-opening): Hermes Agent; Hermes Agent handles this as a draft: From the search results I paste, keep only rows that match my name and a state I have lived in. Say what page I should open next. Do not submit a claim. - [Plan a holiday travel roster](https://aiplaybookgems.com/usecases/claude-cowork-plan-a-holiday-travel-roster): Claude Cowork; Claude Cowork handles this as a draft: Who is coming, which nights, and the beds I said I have. Flag overbooking. Do not email relatives. - [Plan a rail day from timetables I paste](https://aiplaybookgems.com/usecases/chatgpt-agent-plan-a-rail-day-from-timetables-i-paste): ChatGPT Agent; ChatGPT Agent handles this as a draft: Connections with at least the buffer I set. Flag tight transfers. Do not buy tickets. - [Prepare a move inventory from room photos](https://aiplaybookgems.com/usecases/grok-bot-prepare-a-move-inventory-from-room-photos): Grok Bot; Grok Bot handles this as a draft: Room, item, and condition in one line each. Say when a photo is too dark. Do not post the list. - [Reconcile a partner's shared travel fund](https://aiplaybookgems.com/usecases/gemini-reconcile-a-partner-s-shared-travel-fund): Gemini; Gemini handles this as a draft: Match transfers to trip costs I tag. Show the balance. Do not move the fund. - [Sort medical PDFs by visit date](https://aiplaybookgems.com/usecases/perplexity-sort-medical-pdfs-by-visit-date): Perplexity; Perplexity handles this as a draft: Propose filenames. Do not delete. Leave the files in place. - [Spot duplicate pharmacy charges](https://aiplaybookgems.com/usecases/gemini-spot-duplicate-pharmacy-charges): Gemini; Gemini handles this as a draft: Group charges by pharmacy and date. Show pairs that look like the same fill billed twice. Do not call the pharmacy. - [Summarize a homeowners association packet](https://aiplaybookgems.com/usecases/hermes-agent-summarize-a-homeowners-association-packet): Hermes Agent; Hermes Agent handles this as a draft: List votes, dues changes, and deadlines. Quote the page or section. Do not vote or pay. - [Summarize a paper's method in one page](https://aiplaybookgems.com/usecases/gemini-summarize-a-paper-s-method-in-one-page): Gemini; Gemini handles this as a draft: Data, method, and the limit the authors name. Do not add results that are not in the PDF. - [Summarize a policy change for the team](https://aiplaybookgems.com/usecases/n8n-summarize-a-policy-change-for-the-team): n8n; n8n handles this as a draft: What changed, who it affects, and the effective date. Quote the memo. Do not announce it. - [Summarize a regulatory page I pasted](https://aiplaybookgems.com/usecases/gemini-summarize-a-regulatory-page-i-pasted): Gemini; Gemini handles this as a draft: Who it applies to and the effective date if printed. Do not give legal advice. - [Turn a CSV of office requests into themes](https://aiplaybookgems.com/usecases/lindy-turn-a-csv-of-office-requests-into-themes): Lindy; Lindy handles this as a draft: Top five request types and an example row for each. Do not fulfill a request. - [Turn a stack trace into a suspect list](https://aiplaybookgems.com/usecases/github-copilot-turn-a-stack-trace-into-a-suspect-list): GitHub Copilot; GitHub Copilot handles this as a draft: Frames in our code, not the library, first. Do not push a fix. - [Turn a ticket into an implementation note](https://aiplaybookgems.com/usecases/cursor-turn-a-ticket-into-an-implementation-note): Cursor; Cursor handles this as a draft: Acceptance checks from the ticket text. Do not start coding until I say so. - [Write alt text and a caption for a screenshot](https://aiplaybookgems.com/usecases/manus-write-alt-text-and-a-caption-for-a-screenshot): Manus; Manus handles this as a draft: Describe what is visible. No claims about results. Do not publish. - [Fix keyboard and name issues on one screen](https://aiplaybookgems.com/usecases/cursor-accessibility-pass): Cursor; Cursor Agent walks one screen for missing names, keyboard traps, and contrast notes, then patches the clear code issues. - [Route a support ticket in LangGraph and pause before reply](https://aiplaybookgems.com/usecases/langgraph-ticket-router): LangGraph; A small LangGraph app classifies a ticket, drafts a reply, and interrupts for a person before anything is sent. - [Turn a Slack report into a Linear issue draft](https://aiplaybookgems.com/usecases/n8n-linear-drafts): n8n; The n8n agent drafts a Linear issue from a bug report in Slack and shows it to you before anything is filed. - [Implement a small feature that already has acceptance notes](https://aiplaybookgems.com/usecases/codex-cli-small-feature): Codex CLI; Codex CLI builds a feature described as observable behavior, runs the tests you name, and leaves unrelated files alone. - [Review a Devin session before anyone merges](https://aiplaybookgems.com/usecases/devin-review-the-session): Devin; Use Devin's session output as a review packet: commands run, files touched, and tests actually executed. The playbook is for the reviewer, not a second coding agent. - [Build a family seat map request](https://aiplaybookgems.com/usecases/gemini-build-a-family-seat-map-request): Gemini; Gemini handles this as a draft: Who needs to sit together, from the passenger list I write. Do not change the booking. - [Build a rent-increase reply from the lease](https://aiplaybookgems.com/usecases/grok-bot-build-a-rent-increase-reply-from-the-lease): Grok Bot; Grok Bot handles this as a draft: Quote the notice period and the increase. Draft a short reply that asks for the calculation. Leave blanks where the lease is silent. Do not send the reply. - [Compare a refund request to the policy](https://aiplaybookgems.com/usecases/langgraph-compare-a-refund-request-to-the-policy): LangGraph; LangGraph handles this as a draft: Approve-shaped or deny-shaped, with the policy sentence. I decide. Do not issue a refund. - [Compare shipping promises on two carts](https://aiplaybookgems.com/usecases/chatgpt-agent-compare-shipping-promises-on-two-carts): ChatGPT Agent; ChatGPT Agent handles this as a draft: Delivery window and shipping fee from the two pages. Nothing else. Do not place the order. - [Draft a first response for a how-to gap](https://aiplaybookgems.com/usecases/n8n-draft-a-first-response-for-a-how-to-gap): n8n; n8n handles this as a draft: Link the closest article I name and say what it does not cover. Do not send. - [Draft a study-group agenda](https://aiplaybookgems.com/usecases/perplexity-draft-a-study-group-agenda): Perplexity; Perplexity handles this as a draft: Three problems from the list I choose. Do not message the group. - [Draft a timeline for a launch checklist](https://aiplaybookgems.com/usecases/claude-cowork-draft-a-timeline-for-a-launch-checklist): Claude Cowork; Claude Cowork handles this as a draft: Dates I already chose, with the dependency I wrote. Do not announce the launch. - [Draft questions about a bill code](https://aiplaybookgems.com/usecases/claude-draft-questions-about-a-bill-code): Claude; Claude handles this as a draft: The code, the amount, and the question. No legal threat. Do not call. - [File downloaded manuals by appliance](https://aiplaybookgems.com/usecases/claude-cowork-file-downloaded-manuals-by-appliance): Claude Cowork; Claude Cowork handles this as a draft: Propose filenames and a folder for each PDF. Do not delete the originals. Do not throw files away. - [Find dead flags in a config file](https://aiplaybookgems.com/usecases/codex-cli-find-dead-flags-in-a-config-file): Codex CLI; Codex CLI handles this as a draft: Flags with no reference in the search I paste. Do not delete them. - [List tax documents still missing](https://aiplaybookgems.com/usecases/claude-cowork-list-tax-documents-still-missing): Claude Cowork; Claude Cowork handles this as a draft: Compare the checklist I write to the files in the folder. Return only the gaps. Do not file. - [Make a pantry restock from a photo of shelves](https://aiplaybookgems.com/usecases/manus-make-a-pantry-restock-from-a-photo-of-shelves): Manus; Manus handles this as a draft: Items that look low. Say when the photo is unreadable. Do not place a grocery order. - [Make a reunion schedule from a shared doc](https://aiplaybookgems.com/usecases/openclaw-make-a-reunion-schedule-from-a-shared-doc): OpenClaw; OpenClaw handles this as a draft: Times, places, and who is bringing what. Flag blanks. Do not invite anyone. - [Make a source log for a memo](https://aiplaybookgems.com/usecases/manus-make-a-source-log-for-a-memo): Manus; Manus handles this as a draft: URL, what I used it for, and the date on the page if present. Do not invent URLs. - [Prep a one-page leave-behind](https://aiplaybookgems.com/usecases/perplexity-prep-a-one-page-leave-behind): Perplexity; Perplexity handles this as a draft: Problem, approach, and the proof point I already have. No new logos. Do not design or send. - [Propose a series of four posts from one essay](https://aiplaybookgems.com/usecases/claude-propose-a-series-of-four-posts-from-one-essay): Claude; Claude handles this as a draft: Each post one idea from the essay. Do not schedule them. - [Review dependency licenses I exported](https://aiplaybookgems.com/usecases/claude-code-review-dependency-licenses-i-exported): Claude Code; Claude Code handles this as a draft: Which ones are not on the allow list I wrote. Do not add packages. - [Summarize a municipal agenda](https://aiplaybookgems.com/usecases/perplexity-summarize-a-municipal-agenda): Perplexity; Perplexity handles this as a draft: Items and the action requested. Do not submit a public comment. - [Summarize a RFP into a go or no-go brief](https://aiplaybookgems.com/usecases/grok-bot-summarize-a-rfp-into-a-go-or-no-go-brief): Grok Bot; Grok Bot handles this as a draft: Requirements we match from my notes, and gaps. Do not submit the RFP. - [Summarize a tutorial I finished](https://aiplaybookgems.com/usecases/claude-summarize-a-tutorial-i-finished): Claude; Claude handles this as a draft: What I can now do, in my words from the checklist. Do not claim a certificate. - [Summarize a utility outage thread](https://aiplaybookgems.com/usecases/openclaw-summarize-a-utility-outage-thread): OpenClaw; OpenClaw handles this as a draft: What is out, the last update time, and the next promised update. Do not post in the thread. - [Summarize a warranty claim I have not sent](https://aiplaybookgems.com/usecases/claude-cowork-summarize-a-warranty-claim-i-have-not-sent): Claude Cowork; Claude Cowork handles this as a draft: Turn my notes into a timeline and a parts list. End with the sentence the manufacturer still needs. Do not submit the claim. - [Summarize a youth sports schedule](https://aiplaybookgems.com/usecases/gemini-summarize-a-youth-sports-schedule): Gemini; Gemini handles this as a draft: Practices and games in one table. Note conflicts with the family calendar I paste. Do not RSVP. - [Summarize loyalty changes that hit my account](https://aiplaybookgems.com/usecases/openclaw-summarize-loyalty-changes-that-hit-my-account): OpenClaw; OpenClaw handles this as a draft: From the email, what changed and the date. Do not redeem points. - [Turn a discharge note into a home schedule](https://aiplaybookgems.com/usecases/claude-cowork-turn-a-discharge-note-into-a-home-schedule): Claude Cowork; Claude Cowork handles this as a draft: Meds and follow-up dates printed on the note. Do not change the plan. - [Turn a whiteboard photo into a process list](https://aiplaybookgems.com/usecases/zapier-agents-turn-a-whiteboard-photo-into-a-process-list): Zapier Agents; Zapier Agents handles this as a draft: Numbered steps. Mark unreadable words. Do not publish the process. - [Turn interview tape notes into a scene list](https://aiplaybookgems.com/usecases/manus-turn-interview-tape-notes-into-a-scene-list): Manus; Manus handles this as a draft: Scene and the quote. Do not publish the piece. - [Turn irregular freelance income into a monthly average](https://aiplaybookgems.com/usecases/hermes-agent-turn-irregular-freelance-income-into-a-monthly-average): Hermes Agent; Hermes Agent handles this as a draft: Use the deposits I label as income. Show the average and the worst month. Do not give tax advice beyond the arithmetic. # Brief a topic with sources I can open URL: https://aiplaybookgems.com/usecases/perplexity-brief-a-topic-with-sources-i-can-open Agent: Perplexity Category: Research Roles: Founder Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Five claims, each with the URL I gave you or the words not in the packet. Do not cite a page you did not open in the packet. ## Prompt You are working as Perplexity on one playbook. Task: Five claims, each with the URL I gave you or the words not in the packet. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not cite a page you did not open in the packet. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Five claims, each with the URL I gave you or the words not in the packet. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not cite a page you did not open in the packet. # Build a 90-day cash view from two accounts URL: https://aiplaybookgems.com/usecases/claude-cowork-build-a-90-day-cash-view-from-two-accounts Agent: Claude Cowork Category: Data Roles: Founder Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour with exports Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: List expected ins and outs I name. Flag weeks that go negative. Do not invent income. Do not transfer money. ## Prompt You are working as Claude Cowork on one playbook. Task: List expected ins and outs I name. Flag weeks that go negative. Do not invent income. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not transfer money. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: List expected ins and outs I name. Flag weeks that go negative. Do not invent income. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not transfer money. # Build a competitor feature grid from pages I saved URL: https://aiplaybookgems.com/usecases/gemini-build-a-competitor-feature-grid-from-pages-i-saved Agent: Gemini Category: Research Roles: Founder Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Feature present or not stated. Do not guess from memory. ## Prompt You are working as Gemini on one playbook. Task: Feature present or not stated. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not guess from memory. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Feature present or not stated. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not guess from memory. # Build a glossary for a new codebase URL: https://aiplaybookgems.com/usecases/gemini-build-a-glossary-for-a-new-codebase Agent: Gemini Category: Coding Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Terms from the README I paste. Do not edit the repo. ## Prompt You are working as Gemini on one playbook. Task: Terms from the README I paste. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not edit the repo. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Terms from the README I paste. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not edit the repo. # Build a trip skeleton from constraints URL: https://aiplaybookgems.com/usecases/openclaw-build-a-trip-skeleton-from-constraints Agent: OpenClaw Category: Research Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://openclaw.ai/ OpenClaw handles this as a draft: Dates, city, and the rule that I will not check a bag. Propose a day outline with no reservations. Do not book. ## Prompt You are working as OpenClaw on one playbook. Task: Dates, city, and the rule that I will not check a bag. Propose a day outline with no reservations. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not book. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open OpenClaw where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Dates, city, and the rule that I will not check a bag. Propose a day outline with no reservations. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not book. # Build a week of school pickup coverage URL: https://aiplaybookgems.com/usecases/openclaw-build-a-week-of-school-pickup-coverage Agent: OpenClaw Category: Operations Roles: Parent Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://openclaw.ai/ OpenClaw handles this as a draft: From the constraints I list, propose who covers which day. Mark conflicts instead of silently double-booking. Do not text the family. ## Prompt You are working as OpenClaw on one playbook. Task: From the constraints I list, propose who covers which day. Mark conflicts instead of silently double-booking. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not text the family. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open OpenClaw where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the constraints I list, propose who covers which day. Mark conflicts instead of silently double-booking. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not text the family. # Build a weekly support themes memo URL: https://aiplaybookgems.com/usecases/zapier-agents-build-a-weekly-support-themes-memo Agent: Zapier Agents Category: Support Roles: Support lead Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://zapier.com/agents Zapier Agents handles this as a draft: Five themes, each with a ticket id I can open. Do not mail the company. ## Prompt You are working as Zapier Agents on one playbook. Task: Five themes, each with a ticket id I can open. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not mail the company. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Zapier Agents where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Five themes, each with a ticket id I can open. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not mail the company. # Build an account brief before a call URL: https://aiplaybookgems.com/usecases/grok-bot-build-an-account-brief-before-a-call Agent: Grok Bot Category: Sales Roles: Founder Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: From the pages I paste: what they sell, a recent change, and two questions. No invented metrics. Do not email the account. ## Prompt You are working as Grok Bot on one playbook. Task: From the pages I paste: what they sell, a recent change, and two questions. No invented metrics. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not email the account. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the pages I paste: what they sell, a recent change, and two questions. No invented metrics. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not email the account. # Cluster cancellation reasons URL: https://aiplaybookgems.com/usecases/lindy-cluster-cancellation-reasons Agent: Lindy Category: Data Roles: Support lead Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: From the exit notes, five reasons with counts. Do not win them back automatically. ## Prompt You are working as Lindy on one playbook. Task: From the exit notes, five reasons with counts. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not win them back automatically. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the exit notes, five reasons with counts. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not win them back automatically. # Compare three phone plans I already pay for URL: https://aiplaybookgems.com/usecases/hermes-agent-compare-three-phone-plans-i-already-pay-for Agent: Hermes Agent Category: Research Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Forty minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Use only the plan pages I paste. Make a table of price, hotspot, and international. Mark cells you could not find. Do not switch my plan. ## Prompt You are working as Hermes Agent on one playbook. Task: Use only the plan pages I paste. Make a table of price, hotspot, and international. Mark cells you could not find. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not switch my plan. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Use only the plan pages I paste. Make a table of price, hotspot, and international. Mark cells you could not find. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not switch my plan. # Compare three product pages I already opened URL: https://aiplaybookgems.com/usecases/chatgpt-agent-compare-three-product-pages-i-already-opened Agent: ChatGPT Agent Category: Research Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://developers.openai.com/api/docs/guides/tools-computer-use ChatGPT Agent handles this as a draft: Price, warranty, and return window in a table. Leave a cell blank rather than guessing. Do not buy. ## Prompt You are working as ChatGPT Agent on one playbook. Task: Price, warranty, and return window in a table. Leave a cell blank rather than guessing. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not buy. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open ChatGPT Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Price, warranty, and return window in a table. Leave a cell blank rather than guessing. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not buy. # Create a key and code inventory URL: https://aiplaybookgems.com/usecases/hermes-agent-create-a-key-and-code-inventory Agent: Hermes Agent Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: List each lock I describe and where the spare lives. Redact nothing I did not already write. Do not store this anywhere but the file I name. ## Prompt You are working as Hermes Agent on one playbook. Task: List each lock I describe and where the spare lives. Redact nothing I did not already write. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not store this anywhere but the file I name. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: List each lock I describe and where the spare lives. Redact nothing I did not already write. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not store this anywhere but the file I name. # Draft a cancellation for a free trial URL: https://aiplaybookgems.com/usecases/perplexity-draft-a-cancellation-for-a-free-trial Agent: Perplexity Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: The steps on the cancellation page I paste, in order, and the date the charge would hit. Do not click cancel for me. ## Prompt You are working as Perplexity on one playbook. Task: The steps on the cancellation page I paste, in order, and the date the charge would hit. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not click cancel for me. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: The steps on the cancellation page I paste, in order, and the date the charge would hit. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not click cancel for me. # Draft a migration plan for one table URL: https://aiplaybookgems.com/usecases/devin-draft-a-migration-plan-for-one-table Agent: Devin Category: Coding Roles: Software engineer Industries: Software Difficulty: hard Cost bucket: medium Estimated cost: Expect a team seat, an automation platform, or steady model usage. Time saved: An hour Status: unverified Source: https://docs.devin.ai/get-started/devin-intro Devin handles this as a draft: Columns and the backfill question. Do not run SQL. ## Prompt You are working as Devin on one playbook. Task: Columns and the backfill question. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not run SQL. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Devin where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Columns and the backfill question. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not run SQL. # Explain a school device policy to a household URL: https://aiplaybookgems.com/usecases/claude-cowork-explain-a-school-device-policy-to-a-household Agent: Claude Cowork Category: Research Roles: Parent Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: What is allowed, what is not, and the consequence the policy names. Do not contact the school. ## Prompt You are working as Claude Cowork on one playbook. Task: What is allowed, what is not, and the consequence the policy names. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not contact the school. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What is allowed, what is not, and the consequence the policy names. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not contact the school. # Extract quotes from a customer interview URL: https://aiplaybookgems.com/usecases/lindy-extract-quotes-from-a-customer-interview Agent: Lindy Category: Sales Roles: Marketer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: Only sentences they said. Mark the timestamp if I included one. Do not publish. ## Prompt You are working as Lindy on one playbook. Task: Only sentences they said. Mark the timestamp if I included one. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not publish. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Only sentences they said. Mark the timestamp if I included one. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not publish. # Extract vocabulary I marked in a text URL: https://aiplaybookgems.com/usecases/cursor-extract-vocabulary-i-marked-in-a-text Agent: Cursor Category: Research Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://cursor.com/docs/agent/overview Cursor handles this as a draft: Word and the sentence I highlighted. Do not translate beyond that sentence. ## Prompt You are working as Cursor on one playbook. Task: Word and the sentence I highlighted. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not translate beyond that sentence. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Cursor where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Word and the sentence I highlighted. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not translate beyond that sentence. # Line a hospital bill up against public prices URL: https://aiplaybookgems.com/usecases/claude-cowork-line-a-hospital-bill-up-against-public-prices Agent: Claude Cowork Category: Data Roles: Operator Industries: Healthcare Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: About an hour of line-by-line reading Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Build a spreadsheet of the line items on the bill I paste, flag charges that repeat, and list questions for the billing office. Do not call the hospital or send a dispute letter. ## Prompt You are working as Claude Cowork on one playbook. Task: Build a spreadsheet of the line items on the bill I paste, flag charges that repeat, and list questions for the billing office. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not call the hospital or send a dispute letter. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Build a spreadsheet of the line items on the bill I paste, flag charges that repeat, and list questions for the billing office. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not call the hospital or send a dispute letter. # List refill dates for household staples URL: https://aiplaybookgems.com/usecases/gemini-list-refill-dates-for-household-staples Agent: Gemini Category: Operations Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: From the purchase dates I give, suggest the next buy date using the cadence I set. Do not subscribe. ## Prompt You are working as Gemini on one playbook. Task: From the purchase dates I give, suggest the next buy date using the cadence I set. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not subscribe. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the purchase dates I give, suggest the next buy date using the cadence I set. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not subscribe. # Make a fixture from a sanitized log URL: https://aiplaybookgems.com/usecases/cursor-make-a-fixture-from-a-sanitized-log Agent: Cursor Category: Coding Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://cursor.com/docs/agent/overview Cursor handles this as a draft: Replace emails with placeholders. Do not use production data. ## Prompt You are working as Cursor on one playbook. Task: Replace emails with placeholders. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not use production data. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Cursor where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Replace emails with placeholders. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not use production data. # Make a guest checklist for a spare room URL: https://aiplaybookgems.com/usecases/lindy-make-a-guest-checklist-for-a-spare-room Agent: Lindy Category: Operations Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: From my notes, linens, wifi, and the one rule I want on the note. Keep it to one page. Do not text the guest. ## Prompt You are working as Lindy on one playbook. Task: From my notes, linens, wifi, and the one rule I want on the note. Keep it to one page. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not text the guest. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From my notes, linens, wifi, and the one rule I want on the note. Keep it to one page. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not text the guest. # Make a lyrics-free jingle brief URL: https://aiplaybookgems.com/usecases/gemini-make-a-lyrics-free-jingle-brief Agent: Gemini Category: Research Roles: Designer Industries: Agencies Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Mood, length, and the words I will not use. Do not generate a final mix for release. ## Prompt You are working as Gemini on one playbook. Task: Mood, length, and the words I will not use. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not generate a final mix for release. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Mood, length, and the words I will not use. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not generate a final mix for release. # Map entry points before a risky change URL: https://aiplaybookgems.com/usecases/claude-code-map-entry-points-before-a-risky-change Agent: Claude Code Category: Coding Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://code.claude.com/docs/en/overview Claude Code handles this as a draft: Scripts, main files, and the test command in the README. Do not push. ## Prompt You are working as Claude Code on one playbook. Task: Scripts, main files, and the test command in the README. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not push. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Code where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Scripts, main files, and the test command in the README. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not push. # Plan an airport transfer with a time cushion URL: https://aiplaybookgems.com/usecases/chatgpt-agent-plan-an-airport-transfer-with-a-time-cushion Agent: ChatGPT Agent Category: Operations Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://developers.openai.com/api/docs/guides/tools-computer-use ChatGPT Agent handles this as a draft: Leave-by time using the durations I provide. Show the arithmetic. Do not hire a car. ## Prompt You are working as ChatGPT Agent on one playbook. Task: Leave-by time using the durations I provide. Show the arithmetic. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not hire a car. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open ChatGPT Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Leave-by time using the durations I provide. Show the arithmetic. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not hire a car. # Prep an agenda for a 25-minute staff meeting URL: https://aiplaybookgems.com/usecases/lindy-prep-an-agenda-for-a-25-minute-staff-meeting Agent: Lindy Category: Operations Roles: Operator Industries: Agencies Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: Three topics from the doc, each with a decision or not. Do not send the invite. ## Prompt You are working as Lindy on one playbook. Task: Three topics from the doc, each with a decision or not. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send the invite. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Three topics from the doc, each with a decision or not. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send the invite. # Prep an interview guide from a research question URL: https://aiplaybookgems.com/usecases/chatgpt-agent-prep-an-interview-guide-from-a-research-question Agent: ChatGPT Agent Category: Research Roles: Founder Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://developers.openai.com/api/docs/guides/tools-computer-use ChatGPT Agent handles this as a draft: Eight questions. No leading answers. Do not contact participants. ## Prompt You are working as ChatGPT Agent on one playbook. Task: Eight questions. No leading answers. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not contact participants. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open ChatGPT Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Eight questions. No leading answers. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not contact participants. # Prep discovery questions for one industry URL: https://aiplaybookgems.com/usecases/zapier-agents-prep-discovery-questions-for-one-industry Agent: Zapier Agents Category: Sales Roles: Founder Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://zapier.com/agents Zapier Agents handles this as a draft: Eight questions tied to the industry page I paste. Do not contact the prospect. ## Prompt You are working as Zapier Agents on one playbook. Task: Eight questions tied to the industry page I paste. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not contact the prospect. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Zapier Agents where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Eight questions tied to the industry page I paste. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not contact the prospect. # Prep questions before a mortgage recast call URL: https://aiplaybookgems.com/usecases/grok-bot-prep-questions-before-a-mortgage-recast-call Agent: Grok Bot Category: Operations Roles: Operator Industries: Finance Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: From the letter, list the fee, the new payment, and three questions. Do not request the recast. ## Prompt You are working as Grok Bot on one playbook. Task: From the letter, list the fee, the new payment, and three questions. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not request the recast. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the letter, list the fee, the new payment, and three questions. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not request the recast. # Propose chapter titles for a draft URL: https://aiplaybookgems.com/usecases/crewai-propose-chapter-titles-for-a-draft Agent: CrewAI Category: Research Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.crewai.com/introduction CrewAI handles this as a draft: Titles that match sections I marked. Do not rewrite the book. ## Prompt You are working as CrewAI on one playbook. Task: Titles that match sections I marked. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not rewrite the book. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open CrewAI where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Titles that match sections I marked. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not rewrite the book. # Sort a shoebox of contractor invoices URL: https://aiplaybookgems.com/usecases/n8n-sort-a-shoebox-of-contractor-invoices Agent: n8n Category: Data Roles: Finance Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An afternoon of filing Status: unverified Source: https://docs.n8n.io/build/build-and-manage-agents n8n handles this as a draft: Name the vendor, date, amount, and whether the PDF includes an address. Flag invoices missing a date. Do not pay anyone. ## Prompt You are working as n8n on one playbook. Task: Name the vendor, date, amount, and whether the PDF includes an address. Flag invoices missing a date. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not pay anyone. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open n8n where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Name the vendor, date, amount, and whether the PDF includes an address. Flag invoices missing a date. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not pay anyone. # Sort the morning mail into four piles URL: https://aiplaybookgems.com/usecases/openclaw-sort-the-morning-mail-into-four-piles Agent: OpenClaw Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://openclaw.ai/ OpenClaw handles this as a draft: From the inbox I approve, label each thread: reply, waiting, fyi, or ignore. Draft replies only for reply. Do not send mail. ## Prompt You are working as OpenClaw on one playbook. Task: From the inbox I approve, label each thread: reply, waiting, fyi, or ignore. Draft replies only for reply. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send mail. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open OpenClaw where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the inbox I approve, label each thread: reply, waiting, fyi, or ignore. Draft replies only for reply. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send mail. # Summarize a benefits booklet section on therapy URL: https://aiplaybookgems.com/usecases/hermes-agent-summarize-a-benefits-booklet-section-on-therapy Agent: Hermes Agent Category: Research Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Visit limits and whether a referral is mentioned. Do not book therapy. ## Prompt You are working as Hermes Agent on one playbook. Task: Visit limits and whether a referral is mentioned. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not book therapy. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Visit limits and whether a referral is mentioned. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not book therapy. # Summarize a caregiver update URL: https://aiplaybookgems.com/usecases/gemini-summarize-a-caregiver-update Agent: Gemini Category: Operations Roles: Parent Industries: Healthcare Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Changes since yesterday in the note I paste. Do not text the family thread. ## Prompt You are working as Gemini on one playbook. Task: Changes since yesterday in the note I paste. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not text the family thread. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Changes since yesterday in the note I paste. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not text the family thread. # Summarize a student-loan letter URL: https://aiplaybookgems.com/usecases/claude-summarize-a-student-loan-letter Agent: Claude Category: Data Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Balance, payment, and the date something changes. Unknown if the letter does not say. Do not consolidate. ## Prompt You are working as Claude on one playbook. Task: Balance, payment, and the date something changes. Unknown if the letter does not say. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not consolidate. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Balance, payment, and the date something changes. Unknown if the letter does not say. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not consolidate. # Summarize access reviews due this month URL: https://aiplaybookgems.com/usecases/hermes-agent-summarize-access-reviews-due-this-month Agent: Hermes Agent Category: Data Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: System, owner, and due date from the sheet. Do not revoke access. ## Prompt You are working as Hermes Agent on one playbook. Task: System, owner, and due date from the sheet. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not revoke access. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: System, owner, and due date from the sheet. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not revoke access. # Summarize visa checklist items still missing URL: https://aiplaybookgems.com/usecases/manus-summarize-visa-checklist-items-still-missing Agent: Manus Category: Research Roles: Operator Industries: Hospitality Difficulty: medium Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Compare the embassy list I paste to the documents I say I have. Do not submit an application. ## Prompt You are working as Manus on one playbook. Task: Compare the embassy list I paste to the documents I say I have. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not submit an application. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Compare the embassy list I paste to the documents I say I have. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not submit an application. # Triage a support queue into four buckets URL: https://aiplaybookgems.com/usecases/n8n-triage-a-support-queue-into-four-buckets Agent: n8n Category: Support Roles: Support lead Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://docs.n8n.io/build/build-and-manage-agents n8n handles this as a draft: Billing, bug, how-to, and unknown. One sentence why. Do not reply to customers. ## Prompt You are working as n8n on one playbook. Task: Billing, bug, how-to, and unknown. One sentence why. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not reply to customers. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open n8n where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Billing, bug, how-to, and unknown. One sentence why. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not reply to customers. # Turn a chapter into ten questions I can answer URL: https://aiplaybookgems.com/usecases/claude-turn-a-chapter-into-ten-questions-i-can-answer Agent: Claude Category: Research Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Questions from the chapter I paste, with the page line they came from. Do not take the quiz for me. ## Prompt You are working as Claude on one playbook. Task: Questions from the chapter I paste, with the page line they came from. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not take the quiz for me. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Questions from the chapter I paste, with the page line they came from. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not take the quiz for me. # Turn a rough brief into three directions URL: https://aiplaybookgems.com/usecases/claude-turn-a-rough-brief-into-three-directions Agent: Claude Category: Research Roles: Designer Industries: Agencies Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Each direction uses only the constraints in the brief. Do not publish. ## Prompt You are working as Claude on one playbook. Task: Each direction uses only the constraints in the brief. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not publish. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Each direction uses only the constraints in the brief. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not publish. # Turn a visit summary into questions for the next appointment URL: https://aiplaybookgems.com/usecases/claude-turn-a-visit-summary-into-questions-for-the-next-appointment Agent: Claude Category: Research Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Questions only, tied to words in the summary. No new diagnoses. Do not message the clinic. ## Prompt You are working as Claude on one playbook. Task: Questions only, tied to words in the summary. No new diagnoses. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not message the clinic. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Questions only, tied to words in the summary. No new diagnoses. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not message the clinic. # Write a babysitter briefing for one evening URL: https://aiplaybookgems.com/usecases/hermes-agent-write-a-babysitter-briefing-for-one-evening Agent: Hermes Agent Category: Operations Roles: Parent Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Bedtime, allergies, and the show they can watch. One page. Do not contact the sitter. ## Prompt You are working as Hermes Agent on one playbook. Task: Bedtime, allergies, and the show they can watch. One page. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not contact the sitter. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Bedtime, allergies, and the show they can watch. One page. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not contact the sitter. # Write a Monday status from ticket titles URL: https://aiplaybookgems.com/usecases/n8n-write-a-monday-status-from-ticket-titles Agent: n8n Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://docs.n8n.io/build/build-and-manage-agents n8n handles this as a draft: Done, next, and blocked, using only the titles I paste. No invented progress. Do not post in Slack. ## Prompt You are working as n8n on one playbook. Task: Done, next, and blocked, using only the titles I paste. No invented progress. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not post in Slack. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open n8n where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Done, next, and blocked, using only the titles I paste. No invented progress. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not post in Slack. # Build a candidate shortlist without contacting anyone URL: https://aiplaybookgems.com/usecases/grok-bot-talent-scout Agent: Grok Bot Category: Recruiting Roles: Recruiter Industries: Software Difficulty: medium Cost bucket: low Estimated cost: Paid Cursor or SuperGrok access, plus whatever sourcing tools you already pay for. Price not re-quoted here. Time saved: The first pass of a sourcing sprint, often a few hours, when the must-have bar is already written. Status: unverified Source: https://docs.x.ai/grok-bot/use-cases A Grok Bot turns a role description into 20 sourced candidate notes and outreach drafts. The first run is not allowed to message candidates. ## Prompt You are a recruiting researcher. Do not email, InMail, or submit anyone to an ATS. ROLE, MUST_HAVES, NICE_TO_HAVES, and EXCLUDE are pasted below. Find 20 people who match the must-haves. For each: - Link the public profile or page you used. - Quote or paraphrase the evidence for every must-have. If a must-have is not visible, say "not shown" and do not count them as a match. - Drop anyone listed in EXCLUDE. - Draft a 70-word note in my voice sample. Do not invent a personal fact. Close with roles you almost included and why you cut them. ROLE: MUST_HAVES: NICE_TO_HAVES: EXCLUDE: VOICE: ## Steps 1. Write must-haves a stranger could check: Replace taste words with evidence: years, stack, domain, location constraints you are allowed to use. 2. Point the bot at approved sources: Connect the ATS if you want exclusions to be real. Grok Bot's talent notes say to respect source terms and regional rules. 3. Run the prompt on one role: Refuse a list that has no link per person. Twenty unsourced names is not a shortlist. 4. Have a recruiter delete weak rows: The bot does not decide who is worth a conversation. A person does. 5. Add an approval before any outreach tool is connected: Keep messaging off until the shortlist quality is boringly consistent. # Build a clinic visit folder checklist URL: https://aiplaybookgems.com/usecases/claude-cowork-build-a-clinic-visit-folder-checklist Agent: Claude Cowork Category: Operations Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: ID, insurance card, and the questions doc. Do not upload them anywhere. ## Prompt You are working as Claude Cowork on one playbook. Task: ID, insurance card, and the questions doc. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not upload them anywhere. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: ID, insurance card, and the questions doc. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not upload them anywhere. # Build a follow-up list of stale opportunities URL: https://aiplaybookgems.com/usecases/manus-build-a-follow-up-list-of-stale-opportunities Agent: Manus Category: Sales Roles: Founder Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Name, last touch, and a draft sentence. I choose who gets it. Do not send or edit amounts. ## Prompt You are working as Manus on one playbook. Task: Name, last touch, and a draft sentence. I choose who gets it. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send or edit amounts. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Name, last touch, and a draft sentence. I choose who gets it. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send or edit amounts. # Build a gift budget that survives December URL: https://aiplaybookgems.com/usecases/hermes-agent-build-a-gift-budget-that-survives-december Agent: Hermes Agent Category: Operations Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Spread the cap I give you across the people on my list. Show what is left. Do not buy anything. ## Prompt You are working as Hermes Agent on one playbook. Task: Spread the cap I give you across the people on my list. Show what is left. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not buy anything. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Spread the cap I give you across the people on my list. Show what is left. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not buy anything. # Build a move-out cost checklist from the lease URL: https://aiplaybookgems.com/usecases/grok-bot-build-a-move-out-cost-checklist-from-the-lease Agent: Grok Bot Category: Operations Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: List deposits, cleaning clauses, and deadlines. Quote the clause number. Do not message the landlord. ## Prompt You are working as Grok Bot on one playbook. Task: List deposits, cleaning clauses, and deadlines. Quote the clause number. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not message the landlord. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: List deposits, cleaning clauses, and deadlines. Quote the clause number. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not message the landlord. # Categorize a month of spending for a budget URL: https://aiplaybookgems.com/usecases/claude-categorize-a-month-of-spending-for-a-budget Agent: Claude Category: Data Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: One evening Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Use only the categories I provide. Put leftovers in Unsorted with the raw description. Do not move money between accounts. ## Prompt You are working as Claude on one playbook. Task: Use only the categories I provide. Put leftovers in Unsorted with the raw description. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not move money between accounts. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Use only the categories I provide. Put leftovers in Unsorted with the raw description. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not move money between accounts. # Cluster survey comments URL: https://aiplaybookgems.com/usecases/crewai-cluster-survey-comments Agent: CrewAI Category: Data Roles: Marketer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://docs.crewai.com/introduction CrewAI handles this as a draft: Five themes and a quote for each. Do not identify people beyond the comment. ## Prompt You are working as CrewAI on one playbook. Task: Five themes and a quote for each. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not identify people beyond the comment. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open CrewAI where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Five themes and a quote for each. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not identify people beyond the comment. # Compare three flights I pasted URL: https://aiplaybookgems.com/usecases/chatgpt-agent-compare-three-flights-i-pasted Agent: ChatGPT Agent Category: Research Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://developers.openai.com/api/docs/guides/tools-computer-use ChatGPT Agent handles this as a draft: Duration, stops, and bag fee if the text includes it. Blank if it does not. Do not buy a ticket. ## Prompt You are working as ChatGPT Agent on one playbook. Task: Duration, stops, and bag fee if the text includes it. Blank if it does not. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not buy a ticket. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open ChatGPT Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Duration, stops, and bag fee if the text includes it. Blank if it does not. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not buy a ticket. # Compare two after-school programs from their PDFs URL: https://aiplaybookgems.com/usecases/claude-compare-two-after-school-programs-from-their-pdfs Agent: Claude Category: Research Roles: Parent Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Forty minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Price, hours, and pickup rules. Mark missing facts. Do not enroll. ## Prompt You are working as Claude on one playbook. Task: Price, hours, and pickup rules. Mark missing facts. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not enroll. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Price, hours, and pickup rules. Mark missing facts. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not enroll. # Compare two mattress return policies URL: https://aiplaybookgems.com/usecases/claude-compare-two-mattress-return-policies Agent: Claude Category: Research Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Trial length, who pays return shipping, and fees. Quotes only. Do not buy a mattress. ## Prompt You are working as Claude on one playbook. Task: Trial length, who pays return shipping, and fees. Quotes only. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not buy a mattress. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Trial length, who pays return shipping, and fees. Quotes only. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not buy a mattress. # Compare two papers on one question URL: https://aiplaybookgems.com/usecases/gemini-compare-two-papers-on-one-question Agent: Gemini Category: Research Roles: Student Industries: Education Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Method and result, side by side, with the limit each paper states. Do not declare a winner without a quote. ## Prompt You are working as Gemini on one playbook. Task: Method and result, side by side, with the limit each paper states. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not declare a winner without a quote. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Method and result, side by side, with the limit each paper states. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not declare a winner without a quote. # Draft a reply that asks for the missing log URL: https://aiplaybookgems.com/usecases/lindy-draft-a-reply-that-asks-for-the-missing-log Agent: Lindy Category: Support Roles: Support lead Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: Polite, specific about which log, no promised fix. Do not send it. ## Prompt You are working as Lindy on one playbook. Task: Polite, specific about which log, no promised fix. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Polite, specific about which log, no promised fix. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send it. # Draft a short story outline from my premise URL: https://aiplaybookgems.com/usecases/gemini-draft-a-short-story-outline-from-my-premise Agent: Gemini Category: Research Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Forty minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Beats only. Keep my ending if I wrote one. Do not post it. ## Prompt You are working as Gemini on one playbook. Task: Beats only. Keep my ending if I wrote one. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not post it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Beats only. Keep my ending if I wrote one. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not post it. # Explain a function from my notes URL: https://aiplaybookgems.com/usecases/gemini-explain-a-function-from-my-notes Agent: Gemini Category: Research Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Use the definition I wrote. Say if my note is incomplete. Do not browse for a different definition. ## Prompt You are working as Gemini on one playbook. Task: Use the definition I wrote. Say if my note is incomplete. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not browse for a different definition. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Use the definition I wrote. Say if my note is incomplete. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not browse for a different definition. # Explain a medical EOB in plain language URL: https://aiplaybookgems.com/usecases/gemini-explain-a-medical-eob-in-plain-language Agent: Gemini Category: Data Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Rewrite what I owe, what insurance paid, and which code is still pending. Do not guess a code you cannot see. Do not advise treatment. ## Prompt You are working as Gemini on one playbook. Task: Rewrite what I owe, what insurance paid, and which code is still pending. Do not guess a code you cannot see. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not advise treatment. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Rewrite what I owe, what insurance paid, and which code is still pending. Do not guess a code you cannot see. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not advise treatment. # Extract action items from a customer call URL: https://aiplaybookgems.com/usecases/claude-cowork-extract-action-items-from-a-customer-call Agent: Claude Cowork Category: Support Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: What we promised, by when. Separate their requests we did not accept. Do not email the customer. ## Prompt You are working as Claude Cowork on one playbook. Task: What we promised, by when. Separate their requests we did not accept. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not email the customer. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What we promised, by when. Separate their requests we did not accept. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not email the customer. # Extract feature requests without merging them URL: https://aiplaybookgems.com/usecases/claude-extract-feature-requests-without-merging-them Agent: Claude Category: Support Roles: Support lead Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Quote the request. Do not decide the roadmap. Do not tell the customer it is planned. ## Prompt You are working as Claude on one playbook. Task: Quote the request. Do not decide the roadmap. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not tell the customer it is planned. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Quote the request. Do not decide the roadmap. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not tell the customer it is planned. # Find subscriptions that renewed without a receipt URL: https://aiplaybookgems.com/usecases/hermes-agent-find-subscriptions-that-renewed-without-a-receipt Agent: Hermes Agent Category: Data Roles: Finance Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: A sitting with a card export Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: From the card export I provide, list merchants that charged more than once, the cadence, and which ones have no matching receipt in the folder. Do not cancel any subscription. ## Prompt You are working as Hermes Agent on one playbook. Task: From the card export I provide, list merchants that charged more than once, the cadence, and which ones have no matching receipt in the folder. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not cancel any subscription. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the card export I provide, list merchants that charged more than once, the cadence, and which ones have no matching receipt in the folder. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not cancel any subscription. # Have a Zapier Agent prepare CRM follow-ups and wait URL: https://aiplaybookgems.com/usecases/zapier-crm-review-list Agent: Zapier Agents Category: Sales Roles: Sales Industries: Software Difficulty: easy Cost bucket: medium Estimated cost: A Zapier plan that includes Agents, plus the CRM you already connect. Check Zapier's current plan page. Time saved: Writing the first draft of follow-ups for a stale-deal view. Status: unverified Source: https://zapier.com/agents Zapier Agents draft follow-ups for stale deals and park them for approval. The playbook does not let the agent send or edit opportunity amounts. ## Prompt Look at opportunities in the view I name that have had no activity for 14 days. For each, draft a follow-up that references the last activity actually in the CRM. If there is no last activity, say "no activity on file" and do not invent a meeting. Do not send email, change the amount, or move the stage. Deliver a review list with the opportunity link and the draft. I will mark SEND next to the ones that may go out. VIEW: ## Steps 1. Open Zapier Agents and connect the CRM: The product page is zapier.com/agents. Use a user whose CRM permissions cannot edit amounts if you have that role. 2. Paste the prompt and name one saved view: A view you already trust beats a filter the agent invents. 3. Read five drafts against the CRM record: Invented meetings are the failure mode. The prompt tells the agent to admit an empty history. 4. Send the marked rows yourself or with a later approved action: Keep SEND as a human mark. Do not map it to an automatic send on day one. # Hold a week of meetings that lack an agenda URL: https://aiplaybookgems.com/usecases/hermes-agent-hold-a-week-of-meetings-that-lack-an-agenda Agent: Hermes Agent Category: Operations Roles: Operator Industries: Agencies Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Find events this week with an empty description. Draft a one-line question I could ask the organizer. Do not decline or edit the invite. ## Prompt You are working as Hermes Agent on one playbook. Task: Find events this week with an empty description. Draft a one-line question I could ask the organizer. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not decline or edit the invite. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Find events this week with an empty description. Draft a one-line question I could ask the organizer. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not decline or edit the invite. # Let an n8n agent sort a Slack channel into a digest URL: https://aiplaybookgems.com/usecases/n8n-slack-triage-agent Agent: n8n Category: Automation Roles: Operations, Support lead Industries: Agencies Difficulty: medium Cost bucket: low Estimated cost: n8n Cloud or a self-hosted instance, plus the model the agent uses. Agents are documented as a preview. Time saved: Skimming a noisy channel, if the digest stays short enough to trust. Status: unverified Source: https://docs.n8n.io/build/build-and-manage-agents An n8n agent reads one Slack channel, classifies messages, and posts a digest only to you. It does not reply in the channel. ## Prompt You watch one Slack channel. For each new message, classify it as incident, customer, ask, or chatter. Every weekday at 16:00 in the channel's timezone, send me a private digest: - Incidents first, with the permalink - Asks that named me or my team - A count of chatter you ignored Do not reply in the channel. Do not DM anyone except the user who owns this agent. If you are unsure, file the message under asks rather than dropping it. ## Steps 1. Turn agents on and create one in the Agent Builder: n8n's build-and-manage-agents doc says agents can take actions in Slack and that the feature is in preview. An admin can disable agents instance-wide. 2. Connect Slack with the least access that can read one channel: Post the digest to your own DM or a private channel, not back into the source. 3. Put the prompt in the agent instructions: Run it on yesterday's messages before you add the schedule. 4. Schedule only after a week of manual runs: Preview software can change. Keep a human scan for real incidents until the classes look right. # List vaccination record gaps URL: https://aiplaybookgems.com/usecases/gemini-list-vaccination-record-gaps Agent: Gemini Category: Data Roles: Parent Industries: Healthcare Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Compare the record I upload to the school's required list. Return only missing lines. Do not book a clinic. ## Prompt You are working as Gemini on one playbook. Task: Compare the record I upload to the school's required list. Return only missing lines. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not book a clinic. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Compare the record I upload to the school's required list. Return only missing lines. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not book a clinic. # Make a medication list from labels I transcribe URL: https://aiplaybookgems.com/usecases/gemini-make-a-medication-list-from-labels-i-transcribe Agent: Gemini Category: Data Roles: Operator Industries: Healthcare Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Name, dose, and time as printed. Flag conflicts I should ask about, without telling me to stop a medicine. I confirm with a clinician. ## Prompt You are working as Gemini on one playbook. Task: Name, dose, and time as printed. Flag conflicts I should ask about, without telling me to stop a medicine. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: I confirm with a clinician. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Name, dose, and time as printed. Flag conflicts I should ask about, without telling me to stop a medicine. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: I confirm with a clinician. # Organize vaccination dates I typed URL: https://aiplaybookgems.com/usecases/perplexity-organize-vaccination-dates-i-typed Agent: Perplexity Category: Data Roles: Parent Industries: Healthcare Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: A table. Mark missing dates. Do not book a pharmacy. ## Prompt You are working as Perplexity on one playbook. Task: A table. Mark missing dates. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not book a pharmacy. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: A table. Mark missing dates. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not book a pharmacy. # Outline an API change for one field URL: https://aiplaybookgems.com/usecases/codex-cli-outline-an-api-change-for-one-field Agent: Codex CLI Category: Coding Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Forty minutes Status: unverified Source: https://learn.chatgpt.com/docs/codex/cli Codex CLI handles this as a draft: Old, new, and who reads it in the snippets I paste. Do not ship the API. ## Prompt You are working as Codex CLI on one playbook. Task: Old, new, and who reads it in the snippets I paste. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not ship the API. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Codex CLI where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Old, new, and who reads it in the snippets I paste. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not ship the API. # Prep a budget variance note URL: https://aiplaybookgems.com/usecases/grok-bot-prep-a-budget-variance-note Agent: Grok Bot Category: Data Roles: Finance Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: Lines that moved more than the percent I set, with the reason column if present. Do not change the budget. ## Prompt You are working as Grok Bot on one playbook. Task: Lines that moved more than the percent I set, with the reason column if present. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not change the budget. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Lines that moved more than the percent I set, with the reason column if present. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not change the budget. # Prep a lab report outline URL: https://aiplaybookgems.com/usecases/hermes-agent-prep-a-lab-report-outline Agent: Hermes Agent Category: Research Roles: Student Industries: Education Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Sections the rubric requires and which of my notes fill them. Do not write the conclusion as fact. ## Prompt You are working as Hermes Agent on one playbook. Task: Sections the rubric requires and which of my notes fill them. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not write the conclusion as fact. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Sections the rubric requires and which of my notes fill them. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not write the conclusion as fact. # Prep questions for a high-ticket salesperson URL: https://aiplaybookgems.com/usecases/manus-prep-questions-for-a-high-ticket-salesperson Agent: Manus Category: Research Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Five questions from the spec sheet gaps. Do not contact the store. ## Prompt You are working as Manus on one playbook. Task: Five questions from the spec sheet gaps. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not contact the store. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Five questions from the spec sheet gaps. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not contact the store. # Prepare a security report from a user email URL: https://aiplaybookgems.com/usecases/langgraph-prepare-a-security-report-from-a-user-email Agent: LangGraph Category: Support Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://docs.langchain.com/oss/javascript/langgraph/overview LangGraph handles this as a draft: What they saw, when, and what I should not click. Do not reply with a link. ## Prompt You are working as LangGraph on one playbook. Task: What they saw, when, and what I should not click. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not reply with a link. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open LangGraph where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What they saw, when, and what I should not click. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not reply with a link. # Propose a small patch for one failing test URL: https://aiplaybookgems.com/usecases/cursor-propose-a-small-patch-for-one-failing-test Agent: Cursor Category: Coding Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://cursor.com/docs/agent/overview Cursor handles this as a draft: The smallest change that could explain the failure I paste. Do not commit. ## Prompt You are working as Cursor on one playbook. Task: The smallest change that could explain the failure I paste. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not commit. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Cursor where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: The smallest change that could explain the failure I paste. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not commit. # Refactor one module and land tests in the same session URL: https://aiplaybookgems.com/usecases/claude-code-refactor-with-tests Agent: Claude Code Category: Coding Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: A Claude plan that includes Claude Code. Check Anthropic's current plan page. This catalog does not restate prices. Time saved: Several hours on a module you already understand well enough to review. Status: unverified Source: https://code.claude.com/docs/en/overview Claude Code reads a single module, writes characterization tests, then changes structure only where those tests pass. ## Prompt Work only inside the directory I name. Do not touch unrelated packages. 1. Read the module and name its public behavior in a short list. 2. Add characterization tests for that behavior before changing production code. Use the test runner already in the repo. 3. Refactor for readability. Do not change behavior. If a test fails after a refactor, revert that hunk. 4. Show me the diff and the test command output. 5. Stop if the module talks to the network or production data. Ask before stubbing those calls. Done means: tests you added are passing, and the public list still describes the code. ## Steps 1. Install Claude Code and open the repo: Use the current quickstart at code.claude.com. Run it from the repository root so the agent sees the real test script. 2. Name one directory: A whole service is too large for a first refactor. Pick a module with a test runner already configured. 3. Paste the prompt and watch the first test commit: If it starts editing production code before tests exist, stop the session and restate step 2. 4. Read the diff yourself: Passing tests are not a review. Look for behavior hidden in snapshots or deleted assertions. 5. Open a PR with the test command in the body: Claude Code can draft the branch. You merge. # Score a CRM list and stop before anyone is emailed URL: https://aiplaybookgems.com/usecases/grok-bot-account-research Agent: Grok Bot Category: Sales Roles: Sales, Founder Industries: Software Difficulty: medium Cost bucket: low Estimated cost: Included with a paid Cursor plan or a SuperGrok subscription, per xAI's Grok Bot docs. Confirm the current plan matrix before you budget. Time saved: A half day of tab-hopping across a 25-account list, if the sources are already in the CRM export. Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Ask a Grok Bot to score accounts you paste in, name at most two contacts, and draft outreach that stays in a review file. Nothing is sent. ## Prompt You are a sales research bot. Do not email, message, enroll, or edit the CRM. Read three blocks I will paste: ACCOUNTS, ICP, and VOICE. For each account: - Say fit, weak fit, or skip. One sentence of evidence. Name the field or URL you used. - If the public role is obvious, name at most two contacts. Otherwise write "unknown". - Draft a 90-word email and a 40-word note in the VOICE sample. Put [missing] where a fact is not in the sources. - Skip anyone marked already in a sequence. Return a markdown table: account, fit, evidence, contacts, draft. End with accounts you refused because the source was thin. ACCOUNTS: ICP: VOICE: ## Steps 1. Create a bot with a narrow job: In Grok Bot, create a bot whose description is account research only. xAI's use-case docs say to start with read-and-prepare work before approved actions. 2. Connect the CRM as read-only if you can: Prefer an export or a read-only view. Do not grant send rights on the first run. 3. Paste the prompt with a small list: Start with 10 accounts, not 200. Attach the ICP notes and one email you are willing to imitate. 4. Check three rows by hand: Open the cited URL or CRM field. If a contact or fit claim has no source, delete that row's draft. 5. Save the method only after it is dull and correct: Grok Bot docs describe skills and routines as a later step. Save a skill after the review list is reliable, and keep the routine from sending. # Score a lead list against my ICP notes URL: https://aiplaybookgems.com/usecases/zapier-agents-score-a-lead-list-against-my-icp-notes Agent: Zapier Agents Category: Sales Roles: Marketer Industries: Agencies Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://zapier.com/agents Zapier Agents handles this as a draft: Fit or not, with the ICP line you used. Unknown if the row is thin. Do not email leads. ## Prompt You are working as Zapier Agents on one playbook. Task: Fit or not, with the ICP line you used. Unknown if the row is thin. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not email leads. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Zapier Agents where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Fit or not, with the ICP line you used. Unknown if the row is thin. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not email leads. # Sort a downloads folder of life admin PDFs URL: https://aiplaybookgems.com/usecases/claude-cowork-sort-a-downloads-folder-of-life-admin-pdfs Agent: Claude Cowork Category: Data Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Forty minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Propose groups: tax, medical, housing, other. Do not delete. Leave deletion to me. ## Prompt You are working as Claude Cowork on one playbook. Task: Propose groups: tax, medical, housing, other. Do not delete. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Leave deletion to me. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Propose groups: tax, medical, housing, other. Do not delete. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Leave deletion to me. # Sort shared-household expenses for a settle-up URL: https://aiplaybookgems.com/usecases/gemini-sort-shared-household-expenses-for-a-settle-up Agent: Gemini Category: Data Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Forty minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: From the shared sheet, who paid, and the split rule I write. Produce the net. Do not message my housemate. ## Prompt You are working as Gemini on one playbook. Task: From the shared sheet, who paid, and the split rule I write. Produce the net. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not message my housemate. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the shared sheet, who paid, and the split rule I write. Produce the net. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not message my housemate. # Triage school and activity emails URL: https://aiplaybookgems.com/usecases/grok-bot-triage-school-and-activity-emails Agent: Grok Bot Category: Operations Roles: Parent Industries: Education Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: Group by child and by deadline. Surface only items due in seven days. Do not reply to the school. ## Prompt You are working as Grok Bot on one playbook. Task: Group by child and by deadline. Surface only items due in seven days. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not reply to the school. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Group by child and by deadline. Surface only items due in seven days. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not reply to the school. # Triage the inbox from Telegram without auto-sending URL: https://aiplaybookgems.com/usecases/openclaw-inbox-from-telegram Agent: OpenClaw Category: Operations Roles: Founder, Operations Industries: Software Difficulty: medium Cost bucket: free Estimated cost: OpenClaw's site says there is no paid tier for the software. You still pay for the model you connect, and for the machine it runs on. Time saved: The sort step of a morning inbox, often 20–40 minutes, after the mail connection is already working. Status: unverified Source: https://openclaw.ai/ OpenClaw runs on your machine and can meet you in Telegram. This playbook asks it to sort the morning mail and draft replies it is not allowed to send. ## Prompt Look at unread mail from the last 16 hours. Do not send, forward, delete, or archive. Group messages into: needs a decision, waiting on someone else, newsletters, and likely phishing. For needs-a-decision, draft a reply I can edit. Quote the sentence you are answering. If a message asks for money, credentials, or a gift card, put it in likely phishing and draft nothing. Return at most 12 items. End with the count of unread you ignored because they were newsletters. I will tell you in this chat if a draft is allowed to be sent. Until I say "send", you only draft. ## Steps 1. Install the gateway on a machine you control: OpenClaw's docs describe a local gateway and chat channels such as Telegram. Follow the current install guide rather than a copied shell script from this page. 2. Connect mail with the narrowest scope that can read: Read-only first. The project site lists inbox organization as a headline capability. Sending is a separate grant. 3. Pin the prompt as the morning skill: Run it once while you watch. Correct the phishing bucket before you trust the drafts. 4. Keep send as an explicit second message: The prompt refuses to send until you say so in the same chat. Do not 'fix' that on day one. # Turn a broken-appliance note into a parts shortlist URL: https://aiplaybookgems.com/usecases/gemini-turn-a-broken-appliance-note-into-a-parts-shortlist Agent: Gemini Category: Research Roles: Operator Industries: Retail Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Model number and the symptom. List parts that match that symptom on the pages I paste. Do not order parts. ## Prompt You are working as Gemini on one playbook. Task: Model number and the symptom. List parts that match that symptom on the pages I paste. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not order parts. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Model number and the symptom. List parts that match that symptom on the pages I paste. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not order parts. # Turn a camp packet into a packing list and dates URL: https://aiplaybookgems.com/usecases/claude-cowork-turn-a-camp-packet-into-a-packing-list-and-dates Agent: Claude Cowork Category: Operations Roles: Parent Industries: Education Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Dates, forms, and what to pack. Quote the form deadline. Do not register. ## Prompt You are working as Claude Cowork on one playbook. Task: Dates, forms, and what to pack. Quote the form deadline. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not register. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Dates, forms, and what to pack. Quote the form deadline. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not register. # Turn a conference agenda into a personal schedule URL: https://aiplaybookgems.com/usecases/perplexity-turn-a-conference-agenda-into-a-personal-schedule Agent: Perplexity Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Sessions I starred, with conflicts called out. Do not register. ## Prompt You are working as Perplexity on one playbook. Task: Sessions I starred, with conflicts called out. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not register. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Sessions I starred, with conflicts called out. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not register. # Turn a meeting transcript into decisions and owners URL: https://aiplaybookgems.com/usecases/zapier-agents-turn-a-meeting-transcript-into-decisions-and-owners Agent: Zapier Agents Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://zapier.com/agents Zapier Agents handles this as a draft: Decision, owner, and date. Omit small talk. Do not assign work in the tracker. ## Prompt You are working as Zapier Agents on one playbook. Task: Decision, owner, and date. Omit small talk. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not assign work in the tracker. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Zapier Agents where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Decision, owner, and date. Omit small talk. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not assign work in the tracker. # Turn a road-trip doc into daily stops URL: https://aiplaybookgems.com/usecases/gemini-turn-a-road-trip-doc-into-daily-stops Agent: Gemini Category: Research Roles: Operator Industries: Hospitality Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Forty minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Drive legs under the hour cap I set. Flag days that break the cap. Do not reserve lodging. ## Prompt You are working as Gemini on one playbook. Task: Drive legs under the hour cap I set. Flag days that break the cap. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not reserve lodging. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Drive legs under the hour cap I set. Flag days that break the cap. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not reserve lodging. # Turn a ticket into a small Cursor patch URL: https://aiplaybookgems.com/usecases/cursor-ticket-to-patch Agent: Cursor Category: Coding Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: A Cursor subscription that includes Agent. See Cursor's pricing. Not restated here. Time saved: The mechanical edit across a handful of files, after the ticket is specific. Status: unverified Source: https://cursor.com/docs/agent/overview Cursor's agent searches the repo, edits the files for one ticket, and runs the checks you name. The ticket's out-of-scope list is part of the prompt. ## Prompt Implement only the ticket below. Out of scope means do not do it, even if it would be nicer. Search the codebase before editing. Touch the smallest set of files. Match the local style. Run the test command I name. If you cannot run it, say why. In the summary, list files changed, how to verify, and anything you were tempted to refactor but did not. TICKET: OUT OF SCOPE: TEST COMMAND: ## Steps 1. Open the repo in Cursor and start Agent: Cursor's agent overview says Agent can search, edit, and run terminal commands. Use the side pane flow your version ships. 2. Paste a ticket with an explicit out-of-scope list: If the ticket is a paragraph of vibes, rewrite it before you involve the agent. 3. Let it run the test command: Watch the terminal. A summary that claims success without output is incomplete. 4. Review the diff in the editor: Reject drive-by refactors. The prompt asked for a list of temptations so you can see them. # Turn client feedback into a revision list URL: https://aiplaybookgems.com/usecases/manus-turn-client-feedback-into-a-revision-list Agent: Manus Category: Sales Roles: Designer Industries: Agencies Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Each note and the frame it refers to. Do not send the revision. ## Prompt You are working as Manus on one playbook. Task: Each note and the frame it refers to. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send the revision. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Each note and the frame it refers to. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send the revision. # Turn feedback on an essay into a fix list URL: https://aiplaybookgems.com/usecases/perplexity-turn-feedback-on-an-essay-into-a-fix-list Agent: Perplexity Category: Research Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Each comment and the sentence it points at. Do not submit the essay. ## Prompt You are working as Perplexity on one playbook. Task: Each comment and the sentence it points at. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not submit the essay. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Each comment and the sentence it points at. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not submit the essay. # Turn testimonials into usable one-liners URL: https://aiplaybookgems.com/usecases/perplexity-turn-testimonials-into-usable-one-liners Agent: Perplexity Category: Sales Roles: Marketer Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Keep their words. Note if I lack permission to publish. Do not post. ## Prompt You are working as Perplexity on one playbook. Task: Keep their words. Note if I lack permission to publish. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not post. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Keep their words. Note if I lack permission to publish. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not post. # Turn user interviews into jobs-to-be-done lines URL: https://aiplaybookgems.com/usecases/manus-turn-user-interviews-into-jobs-to-be-done-lines Agent: Manus Category: Research Roles: Founder Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: When, I want to, so I can. Use their verbs. Do not roadmap it. ## Prompt You are working as Manus on one playbook. Task: When, I want to, so I can. Use their verbs. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not roadmap it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: When, I want to, so I can. Use their verbs. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not roadmap it. # Write a photo caption that states only what is visible URL: https://aiplaybookgems.com/usecases/claude-cowork-write-a-photo-caption-that-states-only-what-is-visible Agent: Claude Cowork Category: Sales Roles: Designer Industries: Agencies Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: No story the pixels do not show. Do not post. ## Prompt You are working as Claude Cowork on one playbook. Task: No story the pixels do not show. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not post. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: No story the pixels do not show. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not post. # Write a PR description from a diff summary URL: https://aiplaybookgems.com/usecases/github-copilot-write-a-pr-description-from-a-diff-summary Agent: GitHub Copilot Category: Coding Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://docs.github.com/en/copilot/get-started/what-is-github-copilot GitHub Copilot handles this as a draft: What changed and how to test. Do not open the PR. ## Prompt You are working as GitHub Copilot on one playbook. Task: What changed and how to test. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not open the PR. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open GitHub Copilot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What changed and how to test. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not open the PR. # A morning digest that only keeps items tied to written priorities URL: https://aiplaybookgems.com/usecases/grok-bot-morning-digest Agent: Grok Bot Category: Operations Roles: Founder, Operations Industries: Agencies Difficulty: easy Cost bucket: low Estimated cost: A paid Grok Bot plan. The digest itself should be a short run, not an open-ended computer session. Time saved: The first 20–30 minutes of inbox and chat scanning on a busy morning. Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot reads the channels you approve and returns a short list mapped to a priorities doc. It does not send messages or move meetings. ## Prompt Review what changed since yesterday in the channels, inbox, and calendar I have approved. Ignore everything else. PRIORITIES is a short document I paste below. Return only items that map to one of those priorities. For each item: source, one sentence on why it matters, a proposed next step, and whether I owe a decision. Maximum eight items. If nothing qualifies, say so. Do not send messages, decline meetings, or edit the calendar. End with three things you left out because they looked urgent but did not match PRIORITIES. PRIORITIES: ## Steps 1. Write five priorities, not a mission statement: Name the projects or accounts that may interrupt the day. A vague 'grow the company' line makes the digest useless. 2. Approve read access one channel at a time: Grok Bot docs describe a chief-of-staff style digest across Slack, email, calendar, and notes. Start with one inbox. 3. Run it once while you still remember yesterday: Mark each item useful or noise. That correction is the product, not the first draft. 4. Schedule it only after the noise drops: Use a routine at a time you can actually read. Leave sends and calendar edits behind approval. # A vendor memo from public pages, delivered as a file URL: https://aiplaybookgems.com/usecases/manus-vendor-memo Agent: Manus Category: Research Roles: Founder, Product manager Industries: Agencies Difficulty: medium Cost bucket: medium Estimated cost: A Manus plan. The welcome doc points at plans and pricing. This catalog does not copy those numbers. Time saved: A first-draft comparison that would otherwise take a day of browsing, before a person checks every citation. Status: unverified Source: https://manus.im/docs/introduction/welcome Manus is documented as an agent with its own computer. This playbook asks for a sourced vendor memo and a file, not a chatty opinion. ## Prompt Compare the three vendors I name for the job I describe. Use their public pricing, docs, and security pages. Deliver a file with: - A table of what each page actually says about price, deployment, and support - The URL under every cell - A short recommendation that restates the constraint in my JOB line - Open questions, including prices that are "contact sales" Do not create accounts, start trials, or email the vendors. If a page will not load, mark the cell unknown. VENDORS: JOB: ## Steps 1. Start a Manus task with the three vendor names: The welcome doc describes Manus as an agent that takes action on its own computer and delivers a result. Ask for a file up front. 2. Forbid account creation in the prompt: Trials and demo forms are where unattended agents wander. Keep this run on public pages. 3. Open every cited URL: A memo with dead or mismatched links is not ready for a buying meeting. 4. Write the decision yourself: The recommendation section is a draft that must echo your JOB constraint, not a new strategy. # Audit annual card fees against the fee schedule URL: https://aiplaybookgems.com/usecases/claude-cowork-audit-annual-card-fees-against-the-fee-schedule Agent: Claude Cowork Category: Data Roles: Finance Industries: Finance Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Under an hour Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Match each fee on the statement to a line in the schedule I paste. List fees with no match. Do not dispute a charge. ## Prompt You are working as Claude Cowork on one playbook. Task: Match each fee on the statement to a line in the schedule I paste. List fees with no match. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not dispute a charge. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Match each fee on the statement to a line in the schedule I paste. List fees with no match. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not dispute a charge. # Build a gift shortlist under a cap URL: https://aiplaybookgems.com/usecases/chatgpt-agent-build-a-gift-shortlist-under-a-cap Agent: ChatGPT Agent Category: Research Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://developers.openai.com/api/docs/guides/tools-computer-use ChatGPT Agent handles this as a draft: Three options under the cap from pages I provide, with the price on the page. Do not check out. ## Prompt You are working as ChatGPT Agent on one playbook. Task: Three options under the cap from pages I provide, with the price on the page. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not check out. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open ChatGPT Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Three options under the cap from pages I provide, with the price on the page. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not check out. # Build a two-week study plan from a syllabus URL: https://aiplaybookgems.com/usecases/perplexity-build-a-two-week-study-plan-from-a-syllabus Agent: Perplexity Category: Research Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Dates and readings on the syllabus only. Do not email the instructor. ## Prompt You are working as Perplexity on one playbook. Task: Dates and readings on the syllabus only. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not email the instructor. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Dates and readings on the syllabus only. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not email the instructor. # Check a brokerage statement for fees URL: https://aiplaybookgems.com/usecases/grok-bot-check-a-brokerage-statement-for-fees Agent: Grok Bot Category: Data Roles: Finance Industries: Finance Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: List advisory fees and fund expense ratios printed on the statement. Skip tickers you cannot see. Do not trade. ## Prompt You are working as Grok Bot on one playbook. Task: List advisory fees and fund expense ratios printed on the statement. Skip tickers you cannot see. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not trade. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: List advisory fees and fund expense ratios printed on the statement. Skip tickers you cannot see. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not trade. # Check a README's commands against the repo URL: https://aiplaybookgems.com/usecases/claude-code-check-a-readme-s-commands-against-the-repo Agent: Claude Code Category: Coding Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://code.claude.com/docs/en/overview Claude Code handles this as a draft: Which commands are missing a script entry. Do not rewrite the README until I agree. ## Prompt You are working as Claude Code on one playbook. Task: Which commands are missing a script entry. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not rewrite the README until I agree. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Code where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Which commands are missing a script entry. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not rewrite the README until I agree. # Cluster tickets that share a phrase URL: https://aiplaybookgems.com/usecases/langgraph-cluster-tickets-that-share-a-phrase Agent: LangGraph Category: Data Roles: Support lead Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://docs.langchain.com/oss/javascript/langgraph/overview LangGraph handles this as a draft: Group by the error string I name. Count each group. Do not close tickets. ## Prompt You are working as LangGraph on one playbook. Task: Group by the error string I name. Count each group. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not close tickets. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open LangGraph where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Group by the error string I name. Count each group. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not close tickets. # Draft a handoff note for a teammate out tomorrow URL: https://aiplaybookgems.com/usecases/lindy-draft-a-handoff-note-for-a-teammate-out-tomorrow Agent: Lindy Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: Open loops from the doc I paste and the one password process I describe without writing the password. Do not message them. ## Prompt You are working as Lindy on one playbook. Task: Open loops from the doc I paste and the one password process I describe without writing the password. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not message them. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Open loops from the doc I paste and the one password process I describe without writing the password. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not message them. # Draft a noise complaint with the dates I logged URL: https://aiplaybookgems.com/usecases/openclaw-draft-a-noise-complaint-with-the-dates-i-logged Agent: OpenClaw Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://openclaw.ai/ OpenClaw handles this as a draft: Turn the log into a polite note with dates and times. No insults. Do not send it. ## Prompt You are working as OpenClaw on one playbook. Task: Turn the log into a polite note with dates and times. No insults. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open OpenClaw where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Turn the log into a polite note with dates and times. No insults. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send it. # Draft a pet-sitter sheet URL: https://aiplaybookgems.com/usecases/lindy-draft-a-pet-sitter-sheet Agent: Lindy Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: Vet, food, and the emergency number I provide. One page. Do not text the sitter. ## Prompt You are working as Lindy on one playbook. Task: Vet, food, and the emergency number I provide. One page. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not text the sitter. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Vet, food, and the emergency number I provide. One page. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not text the sitter. # Draft a podcast rundown URL: https://aiplaybookgems.com/usecases/claude-draft-a-podcast-rundown Agent: Claude Category: Sales Roles: Marketer Industries: Agencies Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Segments and the question for each guest I named. Do not invite guests. ## Prompt You are working as Claude on one playbook. Task: Segments and the question for each guest I named. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not invite guests. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Segments and the question for each guest I named. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not invite guests. # Draft a portal message asking for a record URL: https://aiplaybookgems.com/usecases/claude-draft-a-portal-message-asking-for-a-record Agent: Claude Category: Operations Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: What record, what dates. Under 80 words. Do not send the portal message. ## Prompt You are working as Claude on one playbook. Task: What record, what dates. Under 80 words. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send the portal message. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What record, what dates. Under 80 words. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send the portal message. # Draft a price-match request URL: https://aiplaybookgems.com/usecases/manus-draft-a-price-match-request Agent: Manus Category: Operations Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Quote the competitor price from the page I paste and the policy line from my store. Do not send it. ## Prompt You are working as Manus on one playbook. Task: Quote the competitor price from the page I paste and the policy line from my store. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Quote the competitor price from the page I paste and the policy line from my store. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send it. # Draft a refund ask from a receipt and a policy URL: https://aiplaybookgems.com/usecases/grok-bot-draft-a-refund-ask-from-a-receipt-and-a-policy Agent: Grok Bot Category: Operations Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes of policy reading Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: Read the receipt and the return policy I paste. Draft the email I would send. Quote the policy line you rely on. Do not send the email. ## Prompt You are working as Grok Bot on one playbook. Task: Read the receipt and the return policy I paste. Draft the email I would send. Quote the policy line you rely on. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send the email. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Read the receipt and the return policy I paste. Draft the email I would send. Quote the policy line you rely on. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send the email. # Draft a renewal risk note URL: https://aiplaybookgems.com/usecases/lindy-draft-a-renewal-risk-note Agent: Lindy Category: Sales Roles: Founder Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: Usage signals I paste and the open ask. Do not email the customer. ## Prompt You are working as Lindy on one playbook. Task: Usage signals I paste and the open ask. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not email the customer. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Usage signals I paste and the open ask. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not email the customer. # Draft a speaker bio from a résumé URL: https://aiplaybookgems.com/usecases/crewai-draft-a-speaker-bio-from-a-r-sum Agent: CrewAI Category: Sales Roles: Founder Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://docs.crewai.com/introduction CrewAI handles this as a draft: 90 words, only jobs on the résumé. Do not publish the bio. ## Prompt You are working as CrewAI on one playbook. Task: 90 words, only jobs on the résumé. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not publish the bio. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open CrewAI where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: 90 words, only jobs on the résumé. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not publish the bio. # Draft a test plan for a bug URL: https://aiplaybookgems.com/usecases/codex-cli-draft-a-test-plan-for-a-bug Agent: Codex CLI Category: Coding Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://learn.chatgpt.com/docs/codex/cli Codex CLI handles this as a draft: Cases, including the one that should fail today. Do not merge. ## Prompt You are working as Codex CLI on one playbook. Task: Cases, including the one that should fail today. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not merge. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Codex CLI where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Cases, including the one that should fail today. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not merge. # Draft commit messages for a split URL: https://aiplaybookgems.com/usecases/devin-draft-commit-messages-for-a-split Agent: Devin Category: Coding Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://docs.devin.ai/get-started/devin-intro Devin handles this as a draft: One message per hunk group I describe. Do not commit. ## Prompt You are working as Devin on one playbook. Task: One message per hunk group I describe. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not commit. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Devin where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: One message per hunk group I describe. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not commit. # Draft five outreach lines I can edit URL: https://aiplaybookgems.com/usecases/perplexity-draft-five-outreach-lines-i-can-edit Agent: Perplexity Category: Sales Roles: Founder Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Each line cites a fact from the page I provide. No sequence that sends itself. Do not send. ## Prompt You are working as Perplexity on one playbook. Task: Each line cites a fact from the page I provide. No sequence that sends itself. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Each line cites a fact from the page I provide. No sequence that sends itself. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send. # Explain a lab PDF in plain words URL: https://aiplaybookgems.com/usecases/claude-cowork-explain-a-lab-pdf-in-plain-words Agent: Claude Cowork Category: Data Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Which values the report itself marks high or low. Do not interpret a normal range you cannot see. Not a diagnosis. ## Prompt You are working as Claude Cowork on one playbook. Task: Which values the report itself marks high or low. Do not interpret a normal range you cannot see. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Not a diagnosis. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Which values the report itself marks high or low. Do not interpret a normal range you cannot see. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Not a diagnosis. # Extract numbers from a report into a table URL: https://aiplaybookgems.com/usecases/manus-extract-numbers-from-a-report-into-a-table Agent: Manus Category: Data Roles: Operator Industries: Finance Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Figure, page, and the unit. Skip charts you cannot read. Do not update the live dashboard. ## Prompt You are working as Manus on one playbook. Task: Figure, page, and the unit. Skip charts you cannot read. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not update the live dashboard. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Figure, page, and the unit. Skip charts you cannot read. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not update the live dashboard. # Extract what a travel insurance policy actually covers URL: https://aiplaybookgems.com/usecases/manus-extract-what-a-travel-insurance-policy-actually-covers Agent: Manus Category: Research Roles: Operator Industries: Finance Difficulty: medium Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Delays, medical, and cancellations, each with the exclusion I should read. Do not buy the policy. ## Prompt You are working as Manus on one playbook. Task: Delays, medical, and cancellations, each with the exclusion I should read. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not buy the policy. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Delays, medical, and cancellations, each with the exclusion I should read. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not buy the policy. # Flag lifestyle creep against a baseline month URL: https://aiplaybookgems.com/usecases/claude-cowork-flag-lifestyle-creep-against-a-baseline-month Agent: Claude Cowork Category: Data Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Compare this month's categories to the baseline month I choose. Only comment on categories that moved more than 20 percent. Do not shame or editorialize. ## Prompt You are working as Claude Cowork on one playbook. Task: Compare this month's categories to the baseline month I choose. Only comment on categories that moved more than 20 percent. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not shame or editorialize. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Compare this month's categories to the baseline month I choose. Only comment on categories that moved more than 20 percent. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not shame or editorialize. # List debts with the rate I can see URL: https://aiplaybookgems.com/usecases/hermes-agent-list-debts-with-the-rate-i-can-see Agent: Hermes Agent Category: Data Roles: Operator Industries: Finance Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Make a table of balance, rate, and minimum from the statements I upload. Leave rate blank when the PDF hides it. Do not apply for a loan. ## Prompt You are working as Hermes Agent on one playbook. Task: Make a table of balance, rate, and minimum from the statements I upload. Leave rate blank when the PDF hides it. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not apply for a loan. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Make a table of balance, rate, and minimum from the statements I upload. Leave rate blank when the PDF hides it. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not apply for a loan. # Make a QA sheet for drafted replies URL: https://aiplaybookgems.com/usecases/zapier-agents-make-a-qa-sheet-for-drafted-replies Agent: Zapier Agents Category: Support Roles: Support lead Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://zapier.com/agents Zapier Agents handles this as a draft: Check tone, policy quote, and whether a promise slipped in. I send the reply. ## Prompt You are working as Zapier Agents on one playbook. Task: Check tone, policy quote, and whether a promise slipped in. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: I send the reply. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Zapier Agents where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Check tone, policy quote, and whether a promise slipped in. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: I send the reply. # Make a RACI from a project brief URL: https://aiplaybookgems.com/usecases/hermes-agent-make-a-raci-from-a-project-brief Agent: Hermes Agent Category: Operations Roles: Operator Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Only the roles named in the brief. Do not add departments. Do not notify anyone. ## Prompt You are working as Hermes Agent on one playbook. Task: Only the roles named in the brief. Do not add departments. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not notify anyone. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Only the roles named in the brief. Do not add departments. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not notify anyone. # Plan a no-spend family weekend from what we have URL: https://aiplaybookgems.com/usecases/hermes-agent-plan-a-no-spend-family-weekend-from-what-we-have Agent: Hermes Agent Category: Operations Roles: Parent Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Activities from the constraints I give: weather, ages, and budget zero. Do not buy tickets. ## Prompt You are working as Hermes Agent on one playbook. Task: Activities from the constraints I give: weather, ages, and budget zero. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not buy tickets. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Activities from the constraints I give: weather, ages, and budget zero. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not buy tickets. # Point Codex CLI at one failing check URL: https://aiplaybookgems.com/usecases/codex-cli-failing-check Agent: Codex CLI Category: Coding Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: A ChatGPT plan that includes Codex, per OpenAI's current docs. Confirm before you rely on a number from memory. Time saved: The first hypothesis and the mechanical fix, when the failure is already reproduced locally. Status: unverified Source: https://learn.chatgpt.com/docs/codex/cli Codex CLI investigates a single red check, patches it, and shows the command output. It does not rewrite the suite to go green. ## Prompt The check named below is failing. Reproduce it with the command I give you. Find the smallest change that makes that command pass. Do not delete tests, skip flags, or widen retries. If the failure is environmental (missing secret, unavailable service), stop and say so. Do not stub production services. Paste the command output before and after. Summarize the diff in five lines. CHECK: COMMAND: ## Steps 1. Install Codex CLI from the current docs: The manual lives at learn.chatgpt.com/docs/codex/cli. Use that install path rather than a copied one-liner from a chat. 2. Reproduce the failure yourself once: If you cannot, the agent will invent a story. The command in the prompt should be the one you just ran. 3. Let Codex edit and re-run that command: Reject a green result that came from skipping the check. 4. Read the five-line summary against the diff: They should describe the same change. Then open the PR yourself. # Prep a handover for weekend support URL: https://aiplaybookgems.com/usecases/n8n-prep-a-handover-for-weekend-support Agent: n8n Category: Support Roles: Support lead Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://docs.n8n.io/build/build-and-manage-agents n8n handles this as a draft: Open sev issues and the workaround already sent. Do not change severity. ## Prompt You are working as n8n on one playbook. Task: Open sev issues and the workaround already sent. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not change severity. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open n8n where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Open sev issues and the workaround already sent. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not change severity. # Prep a lightning talk outline URL: https://aiplaybookgems.com/usecases/cursor-prep-a-lightning-talk-outline Agent: Cursor Category: Research Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://cursor.com/docs/agent/overview Cursor handles this as a draft: Three beats and the example I already have. Do not record it. ## Prompt You are working as Cursor on one playbook. Task: Three beats and the example I already have. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not record it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Cursor where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Three beats and the example I already have. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not record it. # Prep a reading order for a new domain URL: https://aiplaybookgems.com/usecases/chatgpt-agent-prep-a-reading-order-for-a-new-domain Agent: ChatGPT Agent Category: Research Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://developers.openai.com/api/docs/guides/tools-computer-use ChatGPT Agent handles this as a draft: Start here, then this, from the list I already collected. Do not add books I did not list. ## Prompt You are working as ChatGPT Agent on one playbook. Task: Start here, then this, from the list I already collected. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not add books I did not list. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open ChatGPT Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Start here, then this, from the list I already collected. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not add books I did not list. # Prep a travel medical sheet URL: https://aiplaybookgems.com/usecases/openclaw-prep-a-travel-medical-sheet Agent: OpenClaw Category: Operations Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://openclaw.ai/ OpenClaw handles this as a draft: Allergies and the clinic note I paste. One page. Do not give medical advice. ## Prompt You are working as OpenClaw on one playbook. Task: Allergies and the clinic note I paste. One page. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not give medical advice. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open OpenClaw where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Allergies and the clinic note I paste. One page. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not give medical advice. # Propose two meeting times from chat and wait URL: https://aiplaybookgems.com/usecases/openclaw-calendar-hold Agent: OpenClaw Category: Operations Roles: Operations Industries: Agencies Difficulty: easy Cost bucket: free Estimated cost: No OpenClaw subscription. Calendar access uses the account you already have. Model usage is separate. Time saved: The back-and-forth of finding two free slots, a few minutes per meeting once chat is connected. Status: unverified Source: https://openclaw.ai/ From WhatsApp, Telegram, or another connected channel, OpenClaw reads the calendar and proposes holds. It does not send the invite until you confirm. ## Prompt I need a 25-minute meeting with the person named below. Read my calendar for the next 10 weekdays. Propose exactly two slots that are free, inside 09:00–16:00 in my calendar's timezone, and not adjacent to another meeting. Do not send an invite, email, or chat message to them. Reply to me with the two slots, the conflicts you avoided, and a one-sentence invite I can paste. If both slots disappear before I answer, say so and propose again. Send only if I reply "book the first" or "book the second". PERSON: TOPIC: ## Steps 1. Connect one chat channel and the calendar: OpenClaw's site lists calendar management from chat apps. Use the channel you will actually check. 2. State timezone and working hours in the bot notes: Otherwise the two slots will be technically free and socially wrong. 3. Ask for a real meeting you have not booked: Compare the proposed slots with the calendar yourself on the first run. 4. Book with the confirmation phrase: The prompt treats any other reply as not permission to send. # Summarize a parent-teacher note URL: https://aiplaybookgems.com/usecases/gemini-summarize-a-parent-teacher-note Agent: Gemini Category: Operations Roles: Parent Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: What the teacher asked, by when, and what is just context. Do not reply. ## Prompt You are working as Gemini on one playbook. Task: What the teacher asked, by when, and what is just context. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not reply. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What the teacher asked, by when, and what is just context. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not reply. # Timeline a company from filings I paste URL: https://aiplaybookgems.com/usecases/perplexity-timeline-a-company-from-filings-i-paste Agent: Perplexity Category: Research Roles: Founder Industries: Finance Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Date and event only. Do not infer strategy. ## Prompt You are working as Perplexity on one playbook. Task: Date and event only. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not infer strategy. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Date and event only. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not infer strategy. # Triage a support queue into a list Lindy does not send URL: https://aiplaybookgems.com/usecases/lindy-support-triage Agent: Lindy Category: Support Roles: Support lead Industries: Ecommerce Difficulty: easy Cost bucket: medium Estimated cost: A Lindy plan. The site has pricing. This catalog does not copy it. Time saved: The morning sort of a queue, once the help docs are connected. Status: unverified Source: https://lindy.ai/ Lindy's site lists support-queue triage as a use case. This playbook produces a prioritized list and draft replies that wait for a person. ## Prompt Read the new conversations in the queue I connect. Rank them: angry customer, refund, how-to, spam. Draft a reply only for how-to, and only with a sentence I can trace to the help doc you cite. For refunds and angry customers, draft a handoff note instead of a customer-facing apology. Do not send, refund, or close conversations. Return the ranked list with links. If the help doc does not cover the question, write "no article" and skip the draft. ## Steps 1. Create a Lindy for the support queue: The public site lists 'Triage support queue' among use cases. Connect the help center and the queue it names. 2. Paste the prompt as the standing instructions: Run it on yesterday's queue before today is on fire. 3. Check that refunds were not given a customer draft: Those should be handoff notes. If they are apologies with promises, tighten the instruction and rerun. 4. Send replies from the help desk, not from the agent: The person on the queue owns the send button for this playbook. # Turn a disruption text into next actions URL: https://aiplaybookgems.com/usecases/gemini-turn-a-disruption-text-into-next-actions Agent: Gemini Category: Operations Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: What changed, the deadline to act, and the two options the airline actually offered. Do not accept a voucher. ## Prompt You are working as Gemini on one playbook. Task: What changed, the deadline to act, and the two options the airline actually offered. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not accept a voucher. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What changed, the deadline to act, and the two options the airline actually offered. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not accept a voucher. # Turn a group chat export into decisions only URL: https://aiplaybookgems.com/usecases/openclaw-turn-a-group-chat-export-into-decisions-only Agent: OpenClaw Category: Operations Roles: Parent Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://openclaw.ai/ OpenClaw handles this as a draft: List decisions and open questions. Drop banter. Do not post back to the chat. ## Prompt You are working as OpenClaw on one playbook. Task: List decisions and open questions. Drop banter. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not post back to the chat. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open OpenClaw where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: List decisions and open questions. Drop banter. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not post back to the chat. # Turn a notes app dump into a weekend list URL: https://aiplaybookgems.com/usecases/claude-cowork-turn-a-notes-app-dump-into-a-weekend-list Agent: Claude Cowork Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Extract errands, each with a place if I wrote one. Drop anything that is a wish, not a task. Do not message anyone. ## Prompt You are working as Claude Cowork on one playbook. Task: Extract errands, each with a place if I wrote one. Drop anything that is a wish, not a task. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not message anyone. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Extract errands, each with a place if I wrote one. Drop anything that is a wish, not a task. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not message anyone. # Turn a recipe into a shopping list with substitutions URL: https://aiplaybookgems.com/usecases/perplexity-turn-a-recipe-into-a-shopping-list-with-substitutions Agent: Perplexity Category: Operations Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Flag ingredients I said I will not buy and propose one substitute from my pantry list. Do not order groceries. ## Prompt You are working as Perplexity on one playbook. Task: Flag ingredients I said I will not buy and propose one substitute from my pantry list. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not order groceries. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Flag ingredients I said I will not buy and propose one substitute from my pantry list. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not order groceries. # Turn a wearable export into a one-week note URL: https://aiplaybookgems.com/usecases/hermes-agent-turn-a-wearable-export-into-a-one-week-note Agent: Hermes Agent Category: Data Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Sleep and steps totals that are in the file. Skip metrics the file lacks. Not medical advice. ## Prompt You are working as Hermes Agent on one playbook. Task: Sleep and steps totals that are in the file. Skip metrics the file lacks. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Not medical advice. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Sleep and steps totals that are in the file. Skip metrics the file lacks. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Not medical advice. # Turn a webinar transcript into a recap post URL: https://aiplaybookgems.com/usecases/grok-bot-turn-a-webinar-transcript-into-a-recap-post Agent: Grok Bot Category: Sales Roles: Marketer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: Three points and a question. No attendance numbers I did not give. Do not post. ## Prompt You are working as Grok Bot on one playbook. Task: Three points and a question. No attendance numbers I did not give. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not post. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Three points and a question. No attendance numbers I did not give. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not post. # Turn bank alerts into a weekly money memo URL: https://aiplaybookgems.com/usecases/n8n-turn-bank-alerts-into-a-weekly-money-memo Agent: n8n Category: Operations Roles: Founder Industries: Finance Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes a week Status: unverified Source: https://docs.n8n.io/build/build-and-manage-agents n8n handles this as a draft: Cluster alerts by merchant. Lead with anything over the threshold I set. Ignore marketing mail. Do not move money. ## Prompt You are working as n8n on one playbook. Task: Cluster alerts by merchant. Lead with anything over the threshold I set. Ignore marketing mail. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not move money. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open n8n where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Cluster alerts by merchant. Lead with anything over the threshold I set. Ignore marketing mail. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not move money. # Turn office-hour notes into a concept map URL: https://aiplaybookgems.com/usecases/claude-turn-office-hour-notes-into-a-concept-map Agent: Claude Category: Research Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Only terms I wrote down. Do not add chapters we have not reached. ## Prompt You are working as Claude on one playbook. Task: Only terms I wrote down. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not add chapters we have not reached. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Only terms I wrote down. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not add chapters we have not reached. # Write a close-the-loop note after an incident URL: https://aiplaybookgems.com/usecases/n8n-write-a-close-the-loop-note-after-an-incident Agent: n8n Category: Operations Roles: Operator Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://docs.n8n.io/build/build-and-manage-agents n8n handles this as a draft: Timeline and the customer impact sentence, from the doc. Do not email customers. ## Prompt You are working as n8n on one playbook. Task: Timeline and the customer impact sentence, from the doc. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not email customers. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open n8n where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Timeline and the customer impact sentence, from the doc. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not email customers. # Write alt lines for a headline I dislike URL: https://aiplaybookgems.com/usecases/manus-write-alt-lines-for-a-headline-i-dislike Agent: Manus Category: Sales Roles: Marketer Industries: Agencies Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Five options in the voice notes I gave. Do not update the site. ## Prompt You are working as Manus on one playbook. Task: Five options in the voice notes I gave. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not update the site. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Five options in the voice notes I gave. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not update the site. # Audit a cart for duplicate subscriptions URL: https://aiplaybookgems.com/usecases/claude-audit-a-cart-for-duplicate-subscriptions Agent: Claude Category: Data Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Point at items that overlap with subscriptions I listed. Do not purchase the cart. ## Prompt You are working as Claude on one playbook. Task: Point at items that overlap with subscriptions I listed. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not purchase the cart. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Point at items that overlap with subscriptions I listed. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not purchase the cart. # Break a video idea into shots URL: https://aiplaybookgems.com/usecases/crewai-break-a-video-idea-into-shots Agent: CrewAI Category: Research Roles: Designer Industries: Agencies Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.crewai.com/introduction CrewAI handles this as a draft: Shot, what is on screen, and the line. No stock I did not name. Do not publish the video. ## Prompt You are working as CrewAI on one playbook. Task: Shot, what is on screen, and the line. No stock I did not name. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not publish the video. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open CrewAI where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Shot, what is on screen, and the line. No stock I did not name. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not publish the video. # Build a medication schedule for a visit URL: https://aiplaybookgems.com/usecases/claude-cowork-build-a-medication-schedule-for-a-visit Agent: Claude Cowork Category: Data Roles: Parent Industries: Healthcare Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Times and doses from the label text I paste. Ask me if two labels disagree. Do not change a dose. ## Prompt You are working as Claude Cowork on one playbook. Task: Times and doses from the label text I paste. Ask me if two labels disagree. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not change a dose. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Times and doses from the label text I paste. Ask me if two labels disagree. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not change a dose. # Build a packing list for a week at home repairs URL: https://aiplaybookgems.com/usecases/hermes-agent-build-a-packing-list-for-a-week-at-home-repairs Agent: Hermes Agent Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Tools and materials named in the estimate I paste. Flag vague lines like miscellaneous. Do not hire anyone. ## Prompt You are working as Hermes Agent on one playbook. Task: Tools and materials named in the estimate I paste. Flag vague lines like miscellaneous. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not hire anyone. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Tools and materials named in the estimate I paste. Flag vague lines like miscellaneous. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not hire anyone. # Build a school-supply list against a store aisle URL: https://aiplaybookgems.com/usecases/perplexity-build-a-school-supply-list-against-a-store-aisle Agent: Perplexity Category: Research Roles: Parent Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Map the teacher's list to item names. Mark anything ambiguous like folder. Do not check out. ## Prompt You are working as Perplexity on one playbook. Task: Map the teacher's list to item names. Mark anything ambiguous like folder. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not check out. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Map the teacher's list to item names. Mark anything ambiguous like folder. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not check out. # Build a symptom timeline from my notes URL: https://aiplaybookgems.com/usecases/hermes-agent-build-a-symptom-timeline-from-my-notes Agent: Hermes Agent Category: Operations Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Date and symptom in my words. Do not suggest a disease. ## Prompt You are working as Hermes Agent on one playbook. Task: Date and symptom in my words. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not suggest a disease. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Date and symptom in my words. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not suggest a disease. # Check a bibliography for missing fields URL: https://aiplaybookgems.com/usecases/hermes-agent-check-a-bibliography-for-missing-fields Agent: Hermes Agent Category: Research Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Which entries lack a year or a title. Do not fetch new papers. ## Prompt You are working as Hermes Agent on one playbook. Task: Which entries lack a year or a title. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not fetch new papers. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Which entries lack a year or a title. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not fetch new papers. # Check a draft for unsourced sentences URL: https://aiplaybookgems.com/usecases/gemini-check-a-draft-for-unsourced-sentences Agent: Gemini Category: Research Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Forty minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: List sentences that state a number or a cause. Do not add sources you did not open. ## Prompt You are working as Gemini on one playbook. Task: List sentences that state a number or a cause. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not add sources you did not open. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: List sentences that state a number or a cause. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not add sources you did not open. # Check accessibility of a component from the markup URL: https://aiplaybookgems.com/usecases/github-copilot-check-accessibility-of-a-component-from-the-markup Agent: GitHub Copilot Category: Coding Roles: Designer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.github.com/en/copilot/get-started/what-is-github-copilot GitHub Copilot handles this as a draft: Name, role, and keyboard notes from the snippet. Do not restyle the app. ## Prompt You are working as GitHub Copilot on one playbook. Task: Name, role, and keyboard notes from the snippet. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not restyle the app. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open GitHub Copilot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Name, role, and keyboard notes from the snippet. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not restyle the app. # Compare two neighborhoods for a short stay URL: https://aiplaybookgems.com/usecases/chatgpt-agent-compare-two-neighborhoods-for-a-short-stay Agent: ChatGPT Agent Category: Research Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://developers.openai.com/api/docs/guides/tools-computer-use ChatGPT Agent handles this as a draft: From the two guides I paste: walk to the station, noise, and groceries. Do not book a stay. ## Prompt You are working as ChatGPT Agent on one playbook. Task: From the two guides I paste: walk to the station, noise, and groceries. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not book a stay. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open ChatGPT Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the two guides I paste: walk to the station, noise, and groceries. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not book a stay. # Draft a paycheck allocation I can edit URL: https://aiplaybookgems.com/usecases/grok-bot-draft-a-paycheck-allocation-i-can-edit Agent: Grok Bot Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: Split the net pay I give you across the buckets I name. Show the remainder. Do not change direct deposit. ## Prompt You are working as Grok Bot on one playbook. Task: Split the net pay I give you across the buckets I name. Show the remainder. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not change direct deposit. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Split the net pay I give you across the buckets I name. Show the remainder. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not change direct deposit. # Draft a retro from ticket comments URL: https://aiplaybookgems.com/usecases/grok-bot-draft-a-retro-from-ticket-comments Agent: Grok Bot Category: Operations Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: What repeated, with links or ids I included. No blame lines. Do not post the retro. ## Prompt You are working as Grok Bot on one playbook. Task: What repeated, with links or ids I included. No blame lines. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not post the retro. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What repeated, with links or ids I included. No blame lines. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not post the retro. # Draft a status page sentence URL: https://aiplaybookgems.com/usecases/lindy-draft-a-status-page-sentence Agent: Lindy Category: Support Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: Impact and the next update time I choose. No cause if we do not know it. Do not post the status. ## Prompt You are working as Lindy on one playbook. Task: Impact and the next update time I choose. No cause if we do not know it. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not post the status. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Impact and the next update time I choose. No cause if we do not know it. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not post the status. # Explain a COBRA offer before the deadline URL: https://aiplaybookgems.com/usecases/gemini-explain-a-cobra-offer-before-the-deadline Agent: Gemini Category: Data Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: State the monthly cost, who is covered, and the date the offer expires, using only the letter. Do not elect coverage. ## Prompt You are working as Gemini on one playbook. Task: State the monthly cost, who is covered, and the date the offer expires, using only the letter. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not elect coverage. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: State the monthly cost, who is covered, and the date the offer expires, using only the letter. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not elect coverage. # Explain a prior-authorization letter URL: https://aiplaybookgems.com/usecases/gemini-explain-a-prior-authorization-letter Agent: Gemini Category: Data Roles: Operator Industries: Healthcare Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: What was denied, the appeal deadline, and the address. Do not appeal for me. ## Prompt You are working as Gemini on one playbook. Task: What was denied, the appeal deadline, and the address. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not appeal for me. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What was denied, the appeal deadline, and the address. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not appeal for me. # List citations I still need for a draft URL: https://aiplaybookgems.com/usecases/gemini-list-citations-i-still-need-for-a-draft Agent: Gemini Category: Research Roles: Student Industries: Education Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Claims in my draft with no footnote. Do not invent sources. ## Prompt You are working as Gemini on one playbook. Task: Claims in my draft with no footnote. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not invent sources. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Claims in my draft with no footnote. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not invent sources. # List contradictions in a set of articles URL: https://aiplaybookgems.com/usecases/chatgpt-agent-list-contradictions-in-a-set-of-articles Agent: ChatGPT Agent Category: Research Roles: Founder Industries: Software Difficulty: medium Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: An hour Status: unverified Source: https://developers.openai.com/api/docs/guides/tools-computer-use ChatGPT Agent handles this as a draft: Claim A, claim B, and the two titles. Do not smooth them into one story. ## Prompt You are working as ChatGPT Agent on one playbook. Task: Claim A, claim B, and the two titles. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not smooth them into one story. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open ChatGPT Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Claim A, claim B, and the two titles. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not smooth them into one story. # Make a bill calendar from due dates URL: https://aiplaybookgems.com/usecases/claude-make-a-bill-calendar-from-due-dates Agent: Claude Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: List the bill, amount, and due day. Mark anything without a due date. Do not enroll in autopay. ## Prompt You are working as Claude on one playbook. Task: List the bill, amount, and due day. Mark anything without a due date. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not enroll in autopay. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: List the bill, amount, and due day. Mark anything without a due date. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not enroll in autopay. # Make a content calendar week from topics I listed URL: https://aiplaybookgems.com/usecases/manus-make-a-content-calendar-week-from-topics-i-listed Agent: Manus Category: Sales Roles: Marketer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Title and the source I must cite. No invented news. Do not schedule posts. ## Prompt You are working as Manus on one playbook. Task: Title and the source I must cite. No invented news. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not schedule posts. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Title and the source I must cite. No invented news. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not schedule posts. # Make a packing list for a climate I name URL: https://aiplaybookgems.com/usecases/manus-make-a-packing-list-for-a-climate-i-name Agent: Manus Category: Operations Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Clothes and documents. Use the weather notes I paste, not a live lookup unless I included one. Do not buy gear. ## Prompt You are working as Manus on one playbook. Task: Clothes and documents. Use the weather notes I paste, not a live lookup unless I included one. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not buy gear. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Clothes and documents. Use the weather notes I paste, not a live lookup unless I included one. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not buy gear. # Make a shared birthday list with budget caps URL: https://aiplaybookgems.com/usecases/hermes-agent-make-a-shared-birthday-list-with-budget-caps Agent: Hermes Agent Category: Operations Roles: Parent Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Person, date, idea, and cap. Skip anyone without a date. Do not buy gifts. ## Prompt You are working as Hermes Agent on one playbook. Task: Person, date, idea, and cap. Skip anyone without a date. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not buy gifts. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Person, date, idea, and cap. Skip anyone without a date. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not buy gifts. # Make flashcards from a lecture outline URL: https://aiplaybookgems.com/usecases/cursor-make-flashcards-from-a-lecture-outline Agent: Cursor Category: Research Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Forty minutes Status: unverified Source: https://cursor.com/docs/agent/overview Cursor handles this as a draft: Term and the definition in the outline. Skip terms without a definition. Do not invent facts. ## Prompt You are working as Cursor on one playbook. Task: Term and the definition in the outline. Skip terms without a definition. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not invent facts. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Cursor where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Term and the definition in the outline. Skip terms without a definition. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not invent facts. # Map a pricing page to my questions URL: https://aiplaybookgems.com/usecases/zapier-agents-map-a-pricing-page-to-my-questions Agent: Zapier Agents Category: Research Roles: Founder Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://zapier.com/agents Zapier Agents handles this as a draft: What is public and what is contact-sales. Do not fill their form. ## Prompt You are working as Zapier Agents on one playbook. Task: What is public and what is contact-sales. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not fill their form. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Zapier Agents where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What is public and what is contact-sales. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not fill their form. # Note the limits in a benchmark writeup URL: https://aiplaybookgems.com/usecases/crewai-note-the-limits-in-a-benchmark-writeup Agent: CrewAI Category: Research Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://docs.crewai.com/introduction CrewAI handles this as a draft: Hardware, date, and what the authors say they did not test. Do not treat it as a purchase recommendation. ## Prompt You are working as CrewAI on one playbook. Task: Hardware, date, and what the authors say they did not test. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not treat it as a purchase recommendation. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open CrewAI where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Hardware, date, and what the authors say they did not test. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not treat it as a purchase recommendation. # Outline a case study from notes URL: https://aiplaybookgems.com/usecases/claude-cowork-outline-a-case-study-from-notes Agent: Claude Cowork Category: Sales Roles: Marketer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Problem, what we did, and the metric I supplied. No extra metrics. Do not get customer approval for me. ## Prompt You are working as Claude Cowork on one playbook. Task: Problem, what we did, and the metric I supplied. No extra metrics. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not get customer approval for me. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Problem, what we did, and the metric I supplied. No extra metrics. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not get customer approval for me. # Prep a home-maintenance log from photos URL: https://aiplaybookgems.com/usecases/lindy-prep-a-home-maintenance-log-from-photos Agent: Lindy Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: For each photo I attach, write the fixture, the visible issue, and whether it looks urgent. Say when a photo is unclear. Do not book a contractor. ## Prompt You are working as Lindy on one playbook. Task: For each photo I attach, write the fixture, the visible issue, and whether it looks urgent. Say when a photo is unclear. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not book a contractor. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: For each photo I attach, write the fixture, the visible issue, and whether it looks urgent. Say when a photo is unclear. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not book a contractor. # Prep a house sitter note before a trip URL: https://aiplaybookgems.com/usecases/perplexity-prep-a-house-sitter-note-before-a-trip Agent: Perplexity Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Plants, bins, and the neighbor's number I provide. Do not text the sitter. ## Prompt You are working as Perplexity on one playbook. Task: Plants, bins, and the neighbor's number I provide. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not text the sitter. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Plants, bins, and the neighbor's number I provide. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not text the sitter. # Prep a second-opinion packet list URL: https://aiplaybookgems.com/usecases/perplexity-prep-a-second-opinion-packet-list Agent: Perplexity Category: Research Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Which records the specialist's page says to bring. Do not request records. ## Prompt You are working as Perplexity on one playbook. Task: Which records the specialist's page says to bring. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not request records. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Which records the specialist's page says to bring. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not request records. # Prepare questions for a benefits enrollment call URL: https://aiplaybookgems.com/usecases/hermes-agent-prepare-questions-for-a-benefits-enrollment-call Agent: Hermes Agent Category: Research Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: A lunch break with the PDF Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: From the plan summary, list deductibles, out-of-network rules, and three questions I should ask before I enroll. Do not enroll me. ## Prompt You are working as Hermes Agent on one playbook. Task: From the plan summary, list deductibles, out-of-network rules, and three questions I should ask before I enroll. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not enroll me. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the plan summary, list deductibles, out-of-network rules, and three questions I should ask before I enroll. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not enroll me. # Propose types for an untyped function URL: https://aiplaybookgems.com/usecases/cursor-propose-types-for-an-untyped-function Agent: Cursor Category: Coding Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Forty minutes Status: unverified Source: https://cursor.com/docs/agent/overview Cursor handles this as a draft: From the tests I paste. Do not change runtime behavior. ## Prompt You are working as Cursor on one playbook. Task: From the tests I paste. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not change runtime behavior. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Cursor where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the tests I paste. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not change runtime behavior. # Review a diff for risky calls URL: https://aiplaybookgems.com/usecases/devin-review-a-diff-for-risky-calls Agent: Devin Category: Coding Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Forty minutes Status: unverified Source: https://docs.devin.ai/get-started/devin-intro Devin handles this as a draft: Network, auth, and migrations. Quote the lines. Do not approve the PR. ## Prompt You are working as Devin on one playbook. Task: Network, auth, and migrations. Quote the lines. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not approve the PR. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Devin where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Network, auth, and migrations. Quote the lines. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not approve the PR. # Sort an ops inbox by SLA risk URL: https://aiplaybookgems.com/usecases/claude-cowork-sort-an-ops-inbox-by-sla-risk Agent: Claude Cowork Category: Operations Roles: Operator Industries: Agencies Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Which threads mention a deadline inside 24 hours. Do not reply. ## Prompt You are working as Claude Cowork on one playbook. Task: Which threads mention a deadline inside 24 hours. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not reply. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Which threads mention a deadline inside 24 hours. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not reply. # Sort hiring-loop feedback into themes URL: https://aiplaybookgems.com/usecases/zapier-agents-sort-hiring-loop-feedback-into-themes Agent: Zapier Agents Category: Recruiting Roles: Recruiter Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://zapier.com/agents Zapier Agents handles this as a draft: Strengths and risks mentioned by more than one interviewer. Do not advance a candidate. ## Prompt You are working as Zapier Agents on one playbook. Task: Strengths and risks mentioned by more than one interviewer. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not advance a candidate. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Zapier Agents where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Strengths and risks mentioned by more than one interviewer. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not advance a candidate. # Summarize a custody calendar change URL: https://aiplaybookgems.com/usecases/claude-summarize-a-custody-calendar-change Agent: Claude Category: Operations Roles: Parent Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: What moved, which week, and the sentence I should confirm with the other parent. Do not send that sentence. ## Prompt You are working as Claude on one playbook. Task: What moved, which week, and the sentence I should confirm with the other parent. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send that sentence. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What moved, which week, and the sentence I should confirm with the other parent. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send that sentence. # Summarize a partner escalation URL: https://aiplaybookgems.com/usecases/claude-summarize-a-partner-escalation Agent: Claude Category: Support Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Ten minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Who is blocked and the last timestamp. Do not escalate further. ## Prompt You are working as Claude on one playbook. Task: Who is blocked and the last timestamp. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not escalate further. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Who is blocked and the last timestamp. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not escalate further. # Tighten a landing paragraph URL: https://aiplaybookgems.com/usecases/gemini-tighten-a-landing-paragraph Agent: Gemini Category: Sales Roles: Marketer Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Shorter, same facts. Show me the cuts. Do not ship the page. ## Prompt You are working as Gemini on one playbook. Task: Shorter, same facts. Show me the cuts. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not ship the page. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Shorter, same facts. Show me the cuts. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not ship the page. # Turn a call transcript into a CRM note URL: https://aiplaybookgems.com/usecases/lindy-turn-a-call-transcript-into-a-crm-note Agent: Lindy Category: Sales Roles: Founder Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: Pain, next step, and date. No fields I did not hear. Do not update the CRM. ## Prompt You are working as Lindy on one playbook. Task: Pain, next step, and date. No fields I did not hear. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not update the CRM. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Pain, next step, and date. No fields I did not hear. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not update the CRM. # Turn a help-center article into a shorter macro URL: https://aiplaybookgems.com/usecases/zapier-agents-turn-a-help-center-article-into-a-shorter-macro Agent: Zapier Agents Category: Support Roles: Support lead Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://zapier.com/agents Zapier Agents handles this as a draft: The steps only, with the article title at the top. Do not publish the macro. ## Prompt You are working as Zapier Agents on one playbook. Task: The steps only, with the article title at the top. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not publish the macro. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Zapier Agents where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: The steps only, with the article title at the top. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not publish the macro. # Turn a project materials list into a store run URL: https://aiplaybookgems.com/usecases/gemini-turn-a-project-materials-list-into-a-store-run Agent: Gemini Category: Operations Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Group by store if I named one. Quantities from my list only. Do not reserve inventory. ## Prompt You are working as Gemini on one playbook. Task: Group by store if I named one. Quantities from my list only. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not reserve inventory. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Group by store if I named one. Quantities from my list only. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not reserve inventory. # Turn a year of utility bills into a rate check URL: https://aiplaybookgems.com/usecases/n8n-turn-a-year-of-utility-bills-into-a-rate-check Agent: n8n Category: Data Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An evening with PDF bills Status: unverified Source: https://docs.n8n.io/build/build-and-manage-agents n8n handles this as a draft: Summarize usage and total paid by month. Point at the month the rate jumped and the line on the bill that explains it. Do not switch providers or submit a form. ## Prompt You are working as n8n on one playbook. Task: Summarize usage and total paid by month. Point at the month the rate jumped and the line on the bill that explains it. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not switch providers or submit a form. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open n8n where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Summarize usage and total paid by month. Point at the month the rate jumped and the line on the bill that explains it. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not switch providers or submit a form. # Weekly home reset from a standing checklist URL: https://aiplaybookgems.com/usecases/grok-bot-weekly-home-reset-from-a-standing-checklist Agent: Grok Bot Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: Compare the checklist to the notes I add this week. Return only the undone items. Do not nag other people in the house. ## Prompt You are working as Grok Bot on one playbook. Task: Compare the checklist to the notes I add this week. Return only the undone items. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not nag other people in the house. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Compare the checklist to the notes I add this week. Return only the undone items. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not nag other people in the house. # Write the rules for a new sinking fund URL: https://aiplaybookgems.com/usecases/gemini-write-the-rules-for-a-new-sinking-fund Agent: Gemini Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Name the goal, monthly amount, and the date it would fill, using the numbers I provide. Do not open the account. ## Prompt You are working as Gemini on one playbook. Task: Name the goal, monthly amount, and the date it would fill, using the numbers I provide. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not open the account. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Name the goal, monthly amount, and the date it would fill, using the numbers I provide. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not open the account. # Draft a pull request summary from the diff URL: https://aiplaybookgems.com/usecases/copilot-pr-summary Agent: GitHub Copilot Category: Coding Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: A GitHub Copilot plan. See GitHub's current plans. Not copied here. Time saved: Ten minutes of PR-body writing, if you were going to describe the diff anyway. Status: unverified Source: https://docs.github.com/en/copilot/get-started/what-is-github-copilot GitHub Copilot writes a PR summary a reviewer can check against the diff: what changed, how to test, and risks. It does not approve the PR. ## Prompt Write a pull request body for the current diff. Sections: - What changed, in complete sentences, file by file only if there are three files or fewer. Otherwise group by behavior. - How to test. Use commands that exist in this repo. - Risk and rollback. - What I did not do. Do not say the change is safe, complete, or ready. Do not invent a test plan I cannot run. If the diff is empty, say so. ## Steps 1. Open the diff in your IDE or on GitHub: Copilot's docs cover the IDE and GitHub.com surfaces. Use whichever one can see this diff. 2. Paste the prompt with the diff in context: A summary from the ticket alone will flatter the author. The diff has to be in the context. 3. Delete any test step you will not run: The prompt forbids invented commands. Enforce that by hand once. 4. Request review from a person: Copilot does not approve the PR in this playbook. # Reconcile a week of expenses without touching payouts URL: https://aiplaybookgems.com/usecases/grok-bot-expense-week Agent: Grok Bot Category: Data Roles: Finance Industries: Finance Difficulty: medium Cost bucket: low Estimated cost: Paid Grok Bot access. Your expense tool is a separate subscription. Time saved: The sorting pass on a weekly close, often one to two hours, when receipts are already in one inbox. Status: unverified Source: https://docs.x.ai/grok-bot/use-cases The bot builds a weekly expense summary, matches receipts it can see, and drafts one follow-up per owner. It does not send or reimburse. ## Prompt Build this week's expense summary from the export I attach and the POLICY text below. Match receipts from the folder or inbox I point you at. For each exception, cite the policy line. If you cannot cite a line, call it "needs a human" rather than a violation. Draft one follow-up per owner. Do not send it. Do not change categories, approvals, or reimbursements. Totals must add back to the export. Show the tie-out at the bottom. POLICY: ## Steps 1. Export one week, not a year: A small file makes the tie-out checkable. Include the policy as text, not a vibe. 2. Give read access to the receipt inbox: Grok Bot's expense example uses the expense system, email, and a spreadsheet. Skip write access. 3. Reject any total that does not tie: If the summary cannot add back to the export, the drafts are not ready. 4. Send follow-ups yourself: Copy the drafts you agree with. Leave reimbursement changes in the finance tool. # Research a public page set and keep a source log URL: https://aiplaybookgems.com/usecases/chatgpt-agent-public-research Agent: ChatGPT Agent Category: Research Roles: Founder Industries: Agencies Difficulty: medium Cost bucket: low Estimated cost: Whatever plan includes the computer-use capable model you call. The linked source is an API guide, so budget for API or product usage you have actually enabled. Time saved: Clicking through a known URL list and pasting quotes into a doc. Status: unverified Source: https://developers.openai.com/api/docs/guides/tools-computer-use A computer-use style agent visits a list of public URLs and returns notes with the page each claim came from. It does not log in. ## Prompt Visit only the URLs I list. Do not create accounts, accept cookies that imply consent to be contacted, or submit forms. For each URL, record: title, the date if the page shows one, three factual sentences, and one limitation of the source. If a URL redirects, record both addresses. If it fails, record the failure. Do not replace a failed page with a search result unless I asked for a backup. Return a source log I can audit in one sitting. URLS: ## Steps 1. Read the computer-use guide before you grant a browser: The source URL for this playbook is the OpenAI API computer-use guide. It is a tool document. Match your setup to that guide or to the product surface you actually have. 2. Prepare the URL list yourself: An agent that chooses its own sources will pad the log. You choose the pages. 3. Run the prompt with logins disabled: No passwords in the prompt. Public pages only. 4. Open two of the URLs and compare sentences: If a sentence is not on the page, drop it. That is the whole audit. # Upgrade one dependency and read the changelog first URL: https://aiplaybookgems.com/usecases/claude-code-dependency-upgrade Agent: Claude Code Category: Coding Roles: Software engineer Industries: Software Difficulty: hard Cost bucket: low Estimated cost: Claude Code on a paid Claude plan. CI minutes are your own. Time saved: The changelog reading and the mechanical codemod, not the production watch afterward. Status: unverified Source: https://code.claude.com/docs/en/quickstart Claude Code summarizes breaking changes for a single dependency, applies the upgrade, and runs the existing tests. Major bumps stay behind a question. ## Prompt Upgrade only the dependency I name. Do not bump unrelated packages to "make the lockfile green" unless the package manager forces it. If it does, list those bumps separately. Before editing: - Read the changelog or release notes between the installed version and the target. - List breaking changes that could touch this repo, with a file path if you can find a call site. - If the target is a new major version, stop and ask me to confirm. Then upgrade, run the test and typecheck scripts in package.json or the equivalent, and paste failures with the first useful stack line. Do not silence types with a blanket ignore. ## Steps 1. Pick one package and a target version: Write both in the prompt. 'Upgrade everything' is how lockfiles become unreviewable. 2. Ask for the breaking-change list first: You can split this into two turns if the model rushes. The list is the point of the session. 3. Let it run the repo's own checks: Claude Code can execute the test script. Read the output rather than trusting a summary that says all green. 4. Ship the upgrade in its own pull request: Keep refactors out of the same diff so a rollback is possible. # Build a glossary from a technical packet URL: https://aiplaybookgems.com/usecases/crewai-build-a-glossary-from-a-technical-packet Agent: CrewAI Category: Research Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Forty minutes Status: unverified Source: https://docs.crewai.com/introduction CrewAI handles this as a draft: Term and the sentence in the packet. Do not add outside definitions. ## Prompt You are working as CrewAI on one playbook. Task: Term and the sentence in the packet. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not add outside definitions. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open CrewAI where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Term and the sentence in the packet. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not add outside definitions. # Build a vendor contact sheet from contracts URL: https://aiplaybookgems.com/usecases/hermes-agent-build-a-vendor-contact-sheet-from-contracts Agent: Hermes Agent Category: Data Roles: Finance Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Name, renewal, and notice period. Blank if the PDF lacks it. Do not email vendors. ## Prompt You are working as Hermes Agent on one playbook. Task: Name, renewal, and notice period. Blank if the PDF lacks it. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not email vendors. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Name, renewal, and notice period. Blank if the PDF lacks it. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not email vendors. # Check a coupon against the exclusions URL: https://aiplaybookgems.com/usecases/manus-check-a-coupon-against-the-exclusions Agent: Manus Category: Research Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Say whether the item I named is excluded, quoting the exclusion. Do not apply the coupon. ## Prompt You are working as Manus on one playbook. Task: Say whether the item I named is excluded, quoting the exclusion. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not apply the coupon. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Say whether the item I named is excluded, quoting the exclusion. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not apply the coupon. # Compare three vendor security pages URL: https://aiplaybookgems.com/usecases/manus-compare-three-vendor-security-pages Agent: Manus Category: Research Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Certifications named on the page. Blank if absent. Do not fill a security form. ## Prompt You are working as Manus on one playbook. Task: Certifications named on the page. Blank if absent. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not fill a security form. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Certifications named on the page. Blank if absent. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not fill a security form. # Compare two clinic instructions I was given URL: https://aiplaybookgems.com/usecases/perplexity-compare-two-clinic-instructions-i-was-given Agent: Perplexity Category: Research Roles: Operator Industries: Healthcare Difficulty: medium Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Where they disagree. Quote both. Do not pick a treatment. ## Prompt You are working as Perplexity on one playbook. Task: Where they disagree. Quote both. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not pick a treatment. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Where they disagree. Quote both. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not pick a treatment. # Compare two competitor homepages URL: https://aiplaybookgems.com/usecases/manus-compare-two-competitor-homepages Agent: Manus Category: Research Roles: Marketer Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Offer, proof, and call to action, quoted. No traffic claims. Do not publish a teardown. ## Prompt You are working as Manus on one playbook. Task: Offer, proof, and call to action, quoted. No traffic claims. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not publish a teardown. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Offer, proof, and call to action, quoted. No traffic claims. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not publish a teardown. # Compare two explanations of the same idea URL: https://aiplaybookgems.com/usecases/hermes-agent-compare-two-explanations-of-the-same-idea Agent: Hermes Agent Category: Research Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Where they agree and the one sentence they do not. Do not pick a side without a quote. ## Prompt You are working as Hermes Agent on one playbook. Task: Where they agree and the one sentence they do not. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not pick a side without a quote. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Where they agree and the one sentence they do not. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not pick a side without a quote. # Compare two SOPs for contradictions URL: https://aiplaybookgems.com/usecases/n8n-compare-two-sops-for-contradictions Agent: n8n Category: Research Roles: Operator Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Forty minutes Status: unverified Source: https://docs.n8n.io/build/build-and-manage-agents n8n handles this as a draft: Side-by-side disagreements. Ignore matching lines. Do not delete an SOP. ## Prompt You are working as n8n on one playbook. Task: Side-by-side disagreements. Ignore matching lines. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not delete an SOP. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open n8n where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Side-by-side disagreements. Ignore matching lines. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not delete an SOP. # Draft a building-management note about a leak URL: https://aiplaybookgems.com/usecases/grok-bot-draft-a-building-management-note-about-a-leak Agent: Grok Bot Category: Operations Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: Use my notes and the photo description. Ask for a time window. Keep it under 120 words. Do not send it. ## Prompt You are working as Grok Bot on one playbook. Task: Use my notes and the photo description. Ask for a time window. Keep it under 120 words. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Use my notes and the photo description. Ask for a time window. Keep it under 120 words. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send it. # Draft a caregiver handoff URL: https://aiplaybookgems.com/usecases/claude-draft-a-caregiver-handoff Agent: Claude Category: Operations Roles: Parent Industries: Healthcare Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Meds, routines, and the number to call, using only my notes. Do not message the caregiver. ## Prompt You are working as Claude on one playbook. Task: Meds, routines, and the number to call, using only my notes. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not message the caregiver. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Meds, routines, and the number to call, using only my notes. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not message the caregiver. # Draft a delay reimbursement note URL: https://aiplaybookgems.com/usecases/gemini-draft-a-delay-reimbursement-note Agent: Gemini Category: Operations Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Timeline from my notes and the rule I paste. Attach list of receipts I name. Do not file the claim. ## Prompt You are working as Gemini on one playbook. Task: Timeline from my notes and the rule I paste. Attach list of receipts I name. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not file the claim. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Timeline from my notes and the rule I paste. Attach list of receipts I name. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not file the claim. # Draft a partner intro I will send myself URL: https://aiplaybookgems.com/usecases/perplexity-draft-a-partner-intro-i-will-send-myself Agent: Perplexity Category: Sales Roles: Founder Industries: Agencies Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Why them, why us, one ask. Under 120 words. Do not send. ## Prompt You are working as Perplexity on one playbook. Task: Why them, why us, one ask. Under 120 words. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Why them, why us, one ask. Under 120 words. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send. # Edit a transcript into a tight script URL: https://aiplaybookgems.com/usecases/claude-cowork-edit-a-transcript-into-a-tight-script Agent: Claude Cowork Category: Sales Roles: Marketer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Cut repeats. Do not add claims. Do not upload it. ## Prompt You are working as Claude Cowork on one playbook. Task: Cut repeats. Do not add claims. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not upload it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Cut repeats. Do not add claims. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not upload it. # Explain a compiler error next to the line URL: https://aiplaybookgems.com/usecases/perplexity-explain-a-compiler-error-next-to-the-line Agent: Perplexity Category: Coding Roles: Student Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: What the message says and the line I paste. Do not rewrite the whole project. ## Prompt You are working as Perplexity on one playbook. Task: What the message says and the line I paste. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not rewrite the whole project. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What the message says and the line I paste. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not rewrite the whole project. # Explain a credit-card promo before it expires URL: https://aiplaybookgems.com/usecases/hermes-agent-explain-a-credit-card-promo-before-it-expires Agent: Hermes Agent Category: Research Roles: Operator Industries: Finance Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: State the promo APR, the expiration, and what the letter says happens next. Quote the sentence. Do not apply for a card. ## Prompt You are working as Hermes Agent on one playbook. Task: State the promo APR, the expiration, and what the letter says happens next. Quote the sentence. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not apply for a card. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: State the promo APR, the expiration, and what the letter says happens next. Quote the sentence. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not apply for a card. # Explain a module to a new teammate URL: https://aiplaybookgems.com/usecases/github-copilot-explain-a-module-to-a-new-teammate Agent: GitHub Copilot Category: Coding Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.github.com/en/copilot/get-started/what-is-github-copilot GitHub Copilot handles this as a draft: What it owns and what it must not import, from the code. Do not refactor it. ## Prompt You are working as GitHub Copilot on one playbook. Task: What it owns and what it must not import, from the code. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not refactor it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open GitHub Copilot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What it owns and what it must not import, from the code. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not refactor it. # Extract renewal dates from insurance mail URL: https://aiplaybookgems.com/usecases/claude-cowork-extract-renewal-dates-from-insurance-mail Agent: Claude Cowork Category: Data Roles: Operator Industries: Finance Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Make a table of policy, carrier, and renewal date. Skip mail that is an advertisement. Do not change a policy. ## Prompt You are working as Claude Cowork on one playbook. Task: Make a table of policy, carrier, and renewal date. Skip mail that is an advertisement. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not change a policy. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Make a table of policy, carrier, and renewal date. Skip mail that is an advertisement. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not change a policy. # Extract reservation numbers from a mail folder URL: https://aiplaybookgems.com/usecases/perplexity-extract-reservation-numbers-from-a-mail-folder Agent: Perplexity Category: Operations Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Hotel, flight, and confirmation code. Skip marketing. Do not modify a booking. ## Prompt You are working as Perplexity on one playbook. Task: Hotel, flight, and confirmation code. Skip marketing. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not modify a booking. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Hotel, flight, and confirmation code. Skip marketing. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not modify a booking. # Find annual charges hiding in a year of CSV rows URL: https://aiplaybookgems.com/usecases/gemini-find-annual-charges-hiding-in-a-year-of-csv-rows Agent: Gemini Category: Data Roles: Finance Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Forty minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: List merchants billed about once a year and the month they hit. Do not cancel them. ## Prompt You are working as Gemini on one playbook. Task: List merchants billed about once a year and the month they hit. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not cancel them. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: List merchants billed about once a year and the month they hit. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not cancel them. # Find tickets waiting on us more than two days URL: https://aiplaybookgems.com/usecases/claude-find-tickets-waiting-on-us-more-than-two-days Agent: Claude Category: Support Roles: Support lead Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: List id, age, and the last agent line. Do not ping the customer. ## Prompt You are working as Claude on one playbook. Task: List id, age, and the last agent line. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not ping the customer. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: List id, age, and the last agent line. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not ping the customer. # List error paths in a handler URL: https://aiplaybookgems.com/usecases/codex-cli-list-error-paths-in-a-handler Agent: Codex CLI Category: Coding Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://learn.chatgpt.com/docs/codex/cli Codex CLI handles this as a draft: What the user sees for each branch. Do not deploy. ## Prompt You are working as Codex CLI on one playbook. Task: What the user sees for each branch. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not deploy. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Codex CLI where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What the user sees for each branch. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not deploy. # Make a food log summary for a dietitian URL: https://aiplaybookgems.com/usecases/claude-cowork-make-a-food-log-summary-for-a-dietitian Agent: Claude Cowork Category: Data Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Foods and times I wrote. No advice. Do not email the dietitian. ## Prompt You are working as Claude Cowork on one playbook. Task: Foods and times I wrote. No advice. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not email the dietitian. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Foods and times I wrote. No advice. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not email the dietitian. # Prep a family emergency card URL: https://aiplaybookgems.com/usecases/gemini-prep-a-family-emergency-card Agent: Gemini Category: Operations Roles: Parent Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Contacts and the meeting place I name. One card, large type. Do not publish it. ## Prompt You are working as Gemini on one playbook. Task: Contacts and the meeting place I name. One card, large type. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not publish it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Contacts and the meeting place I name. One card, large type. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not publish it. # Prep a Schedule C packet from a bank CSV URL: https://aiplaybookgems.com/usecases/gemini-prep-a-schedule-c-packet-from-a-bank-csv Agent: Gemini Category: Data Roles: Founder Industries: Agencies Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half a day of bookkeeping Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Group expenses into categories I name, mark rows you are unsure about, and total only the rows with a category. Do not file a return or connect to tax software. ## Prompt You are working as Gemini on one playbook. Task: Group expenses into categories I name, mark rows you are unsure about, and total only the rows with a category. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not file a return or connect to tax software. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Group expenses into categories I name, mark rows you are unsure about, and total only the rows with a category. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not file a return or connect to tax software. # Reconcile a reimbursement report to receipts URL: https://aiplaybookgems.com/usecases/grok-bot-reconcile-a-reimbursement-report-to-receipts Agent: Grok Bot Category: Data Roles: Finance Industries: Agencies Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour of matching Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: Each report row needs a receipt filename or the word missing. Totals must use only matched rows. Do not submit the report. ## Prompt You are working as Grok Bot on one playbook. Task: Each report row needs a receipt filename or the word missing. Totals must use only matched rows. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not submit the report. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Each report row needs a receipt filename or the word missing. Totals must use only matched rows. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not submit the report. # Sort billing tickets that need finance URL: https://aiplaybookgems.com/usecases/langgraph-sort-billing-tickets-that-need-finance Agent: LangGraph Category: Support Roles: Finance Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://docs.langchain.com/oss/javascript/langgraph/overview LangGraph handles this as a draft: Which ones mention a charge id. Do not refund. ## Prompt You are working as LangGraph on one playbook. Task: Which ones mention a charge id. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not refund. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open LangGraph where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Which ones mention a charge id. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not refund. # Sort wish-list items by actual need URL: https://aiplaybookgems.com/usecases/claude-sort-wish-list-items-by-actual-need Agent: Claude Category: Operations Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Label each item need, nice, or duplicate of something I said I own. Do not purchase. ## Prompt You are working as Claude on one playbook. Task: Label each item need, nice, or duplicate of something I said I own. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not purchase. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Label each item need, nice, or duplicate of something I said I own. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not purchase. # Turn a brand voice note into a checklist URL: https://aiplaybookgems.com/usecases/manus-turn-a-brand-voice-note-into-a-checklist Agent: Manus Category: Research Roles: Designer Industries: Agencies Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Do and don't from my note only. Do not rewrite the guidelines I did not write. ## Prompt You are working as Manus on one playbook. Task: Do and don't from my note only. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not rewrite the guidelines I did not write. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Do and don't from my note only. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not rewrite the guidelines I did not write. # A sourced market scan with the holes left visible URL: https://aiplaybookgems.com/usecases/perplexity-market-scan Agent: Perplexity Category: Research Roles: Researcher, Marketer Industries: Media Difficulty: easy Cost bucket: free Estimated cost: Perplexity's docs include a free starting point and paid API usage. A single scan can stay on the product's ordinary search if that is what you have. Time saved: An afternoon of open-tab research, with the citation check still on you. Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity searches current public sources and returns a scan that separates numbers it found from numbers it did not. ## Prompt Scan the market for the job described below. Use current public sources. Return: - Five suppliers or approaches, each with a source link and the date on the page if shown - What they charge, or "price not public" - Who each one is for, using the source's words - Two facts you expected to find and did not Do not average prices. Do not cite a number you cannot attach to a URL. JOB: ## Steps 1. Open Perplexity or its API with the prompt: The getting-started docs cover the product and the agent API. Use search with citations for this playbook. 2. Name the job narrowly: 'AI tools' returns a magazine list. 'Tools a 12-person support team uses to draft replies' returns something you can check. 3. Click every link that carries a number: If the number is not on the page, delete it. The prompt already asked for that discipline. 4. Keep the 'not found' section: Those holes are the next search, or a reason to stop. # Ask Devin for a pull request from a tightly written issue URL: https://aiplaybookgems.com/usecases/devin-issue-to-pr Agent: Devin Category: Coding Roles: Software engineer Industries: Software Difficulty: hard Cost bucket: medium Estimated cost: A Devin plan. Cognition's pricing changes. Check devin.ai rather than trusting a remembered seat price. Time saved: A first implementation pass on an issue that already has repro steps. Review time does not go to zero. Status: unverified Source: https://docs.devin.ai/get-started/devin-intro Devin picks up one issue and opens a PR. The playbook tells it which repo, which commands, and what not to touch. ## Prompt Work on the issue URL below in the repository I name. Open a pull request when the test command passes. Constraints: - Do not modify CI to weaken a required check. - Do not upgrade dependencies unless the issue says to. - Explain behavior changes in the PR body with the issue link. - If you need a secret that is not already in the environment, stop. ISSUE: REPOSITORY: TEST COMMAND: ## Steps 1. Read Devin's intro and connect the repo: The getting-started docs cover the first session. Connect only the repository named in the prompt. 2. Rewrite the issue until the test command is real: Devin cannot honestly finish an issue that does not say how to know it is done. 3. Let it open the PR, then review like any other contributor: Look at CI changes and lockfiles first. The prompt forbids weakening checks and casual upgrades. 4. Merge yourself: An agent-opened PR is still a PR. Pair it with the regression-test habit if the issue is a bug. # Hand an X thread to a bot and get the brief in the app URL: https://aiplaybookgems.com/usecases/grok-bot-tag-on-x Agent: Grok Bot Category: Research Roles: Marketer Industries: Media Difficulty: easy Cost bucket: low Estimated cost: Requires a Grok account with X connected, per the tag-on-X docs. Usage draws on the Grok Bot allowance. Time saved: A few minutes of copying a thread into another tool, once the bot is already set up. Status: unverified Source: https://docs.x.ai/grok-bot/tag-on-x Tag @bot on a post so Grok Bot picks up the thread. The public reply is only a confirmation. The brief stays in the bot conversation. ## Prompt When I tag you from X, summarize the thread I replied to. Rules: - Use only that post, the post it replies to, and quoted posts, including still photos. - Separate what was said from what it implies. - List claims that need a source outside the thread. - Do not post the summary on X. The on-X reply stays a short confirmation. - Send the brief back in this bot conversation, with bullets I can paste into Slack myself. If the thread is mostly images without text, say what you can see and what you cannot read. ## Steps 1. Connect X and read the first-run notice: The tag-on-X docs say the first tagged request notifies you that tagged work starts automatically and uses Grok Bot usage. 2. Put the prompt in the bot's instructions: You want the same brief shape every time, not a fresh style per tag. 3. Reply to a low-stakes thread with @bot: Ask for a summary. Confirm the public reply is short and the real brief lands in the app. 4. Paste into Slack yourself: This playbook does not grant Slack send. Add that later with an approval if you want it. # Write the regression test for a bug you can reproduce URL: https://aiplaybookgems.com/usecases/cursor-regression-tests Agent: Cursor Category: Coding Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: Cursor Agent on a normal paid plan. Time saved: The boilerplate of a regression test, often under an hour when the repro is already known. Status: unverified Source: https://cursor.com/docs/agent/overview Cursor Agent adds a failing test for a known bug, then the fix, and stops if the test never failed. ## Prompt Here is a bug, the steps that reproduce it, and the file I think is involved. First add a regression test that fails for the right reason. Show me the failure output. If you cannot make it fail, stop and explain. Do not "fix" the production code first. After the failure is real, apply the smallest fix that turns that test green. Do not delete the test or weaken the assertion. BUG: REPRO STEPS: SUSPECT FILE: ## Steps 1. Write the repro in human language first: Include inputs and the wrong output. The agent should not have to rediscover the bug. 2. Start Agent with the prompt: The agent overview lists terminal commands among the tools. The failure output is required. 3. Read the assertion: A test that only checks 'does not throw' is not a regression test for a logic bug. 4. Commit test and fix together: The history should show the test was meaningful. You still review the production change. # Build an on-call handoff URL: https://aiplaybookgems.com/usecases/zapier-agents-build-an-on-call-handoff Agent: Zapier Agents Category: Operations Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://zapier.com/agents Zapier Agents handles this as a draft: Open incidents and the last action, from the notes I paste. Do not page anyone. ## Prompt You are working as Zapier Agents on one playbook. Task: Open incidents and the last action, from the notes I paste. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not page anyone. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Zapier Agents where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Open incidents and the last action, from the notes I paste. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not page anyone. # Cluster campaign results I exported URL: https://aiplaybookgems.com/usecases/lindy-cluster-campaign-results-i-exported Agent: Lindy Category: Data Roles: Marketer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: Which message had the replies, using the columns I name. Do not launch a new campaign. ## Prompt You are working as Lindy on one playbook. Task: Which message had the replies, using the columns I name. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not launch a new campaign. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Which message had the replies, using the columns I name. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not launch a new campaign. # Compare two insurance EOBs for the same visit URL: https://aiplaybookgems.com/usecases/claude-cowork-compare-two-insurance-eobs-for-the-same-visit Agent: Claude Cowork Category: Data Roles: Operator Industries: Healthcare Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour with two PDFs Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Explain what each document says was billed, allowed, and owed. List contradictions instead of averaging them. Do not call the insurer. ## Prompt You are working as Claude Cowork on one playbook. Task: Explain what each document says was billed, allowed, and owed. List contradictions instead of averaging them. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not call the insurer. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Explain what each document says was billed, allowed, and owed. List contradictions instead of averaging them. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not call the insurer. # Compare used versus new from listings I paste URL: https://aiplaybookgems.com/usecases/chatgpt-agent-compare-used-versus-new-from-listings-i-paste Agent: ChatGPT Agent Category: Research Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://developers.openai.com/api/docs/guides/tools-computer-use ChatGPT Agent handles this as a draft: Price, seller type, and return rule. Skip listings without a price. Do not message sellers. ## Prompt You are working as ChatGPT Agent on one playbook. Task: Price, seller type, and return rule. Skip listings without a price. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not message sellers. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open ChatGPT Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Price, seller type, and return rule. Skip listings without a price. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not message sellers. # Draft a charity-donation log for taxes URL: https://aiplaybookgems.com/usecases/n8n-draft-a-charity-donation-log-for-taxes Agent: n8n Category: Data Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://docs.n8n.io/build/build-and-manage-agents n8n handles this as a draft: From the confirmation emails I paste, make a table of date, organization, and amount. Skip anything without an amount. Do not file taxes. ## Prompt You are working as n8n on one playbook. Task: From the confirmation emails I paste, make a table of date, organization, and amount. Skip anything without an amount. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not file taxes. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open n8n where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the confirmation emails I paste, make a table of date, organization, and amount. Skip anything without an amount. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not file taxes. # Draft a hotel request for an early check-in URL: https://aiplaybookgems.com/usecases/openclaw-draft-a-hotel-request-for-an-early-check-in Agent: OpenClaw Category: Operations Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://openclaw.ai/ OpenClaw handles this as a draft: One short note. Mention the arrival time I give. Do not send it. ## Prompt You are working as OpenClaw on one playbook. Task: One short note. Mention the arrival time I give. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open OpenClaw where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: One short note. Mention the arrival time I give. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send it. # Draft a launch email from a changelog URL: https://aiplaybookgems.com/usecases/grok-bot-draft-a-launch-email-from-a-changelog Agent: Grok Bot Category: Sales Roles: Marketer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: Three bullets from the changelog only. Do not send the email. ## Prompt You are working as Grok Bot on one playbook. Task: Three bullets from the changelog only. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send the email. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Three bullets from the changelog only. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send the email. # Draft a rollback note URL: https://aiplaybookgems.com/usecases/devin-draft-a-rollback-note Agent: Devin Category: Coding Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://docs.devin.ai/get-started/devin-intro Devin handles this as a draft: The command in the runbook I paste and the signal that means stop. Do not execute it. ## Prompt You are working as Devin on one playbook. Task: The command in the runbook I paste and the signal that means stop. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not execute it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Devin where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: The command in the runbook I paste and the signal that means stop. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not execute it. # Draft a spending pause for one category URL: https://aiplaybookgems.com/usecases/grok-bot-draft-a-spending-pause-for-one-category Agent: Grok Bot Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: Propose a 30-day rule for the category I name, using last month's actual total. Do not lock a card. ## Prompt You are working as Grok Bot on one playbook. Task: Propose a 30-day rule for the category I name, using last month's actual total. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not lock a card. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Propose a 30-day rule for the category I name, using last month's actual total. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not lock a card. # Extract form fields from a permission slip URL: https://aiplaybookgems.com/usecases/openclaw-extract-form-fields-from-a-permission-slip Agent: OpenClaw Category: Operations Roles: Parent Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://openclaw.ai/ OpenClaw handles this as a draft: List every blank and the fact I still need to supply. Do not sign. ## Prompt You are working as OpenClaw on one playbook. Task: List every blank and the fact I still need to supply. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not sign. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open OpenClaw where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: List every blank and the fact I still need to supply. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not sign. # List upgrade steps from a changelog URL: https://aiplaybookgems.com/usecases/claude-code-list-upgrade-steps-from-a-changelog Agent: Claude Code Category: Coding Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Forty minutes Status: unverified Source: https://code.claude.com/docs/en/overview Claude Code handles this as a draft: Breaking changes that mention the version I am on. Do not run the upgrade. ## Prompt You are working as Claude Code on one playbook. Task: Breaking changes that mention the version I am on. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not run the upgrade. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Code where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Breaking changes that mention the version I am on. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not run the upgrade. # Make a group dinner shortlist URL: https://aiplaybookgems.com/usecases/manus-make-a-group-dinner-shortlist Agent: Manus Category: Research Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Three places from the links I provide, with the reservation rule on the page. Do not book a table. ## Prompt You are working as Manus on one playbook. Task: Three places from the links I provide, with the reservation rule on the page. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not book a table. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Three places from the links I provide, with the reservation rule on the page. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not book a table. # Make a moodboard checklist URL: https://aiplaybookgems.com/usecases/claude-make-a-moodboard-checklist Agent: Claude Category: Research Roles: Designer Industries: Agencies Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Colors and references I listed, grouped. Do not download images. ## Prompt You are working as Claude on one playbook. Task: Colors and references I listed, grouped. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not download images. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Colors and references I listed, grouped. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not download images. # Make a reading schedule around my shifts URL: https://aiplaybookgems.com/usecases/cursor-make-a-reading-schedule-around-my-shifts Agent: Cursor Category: Operations Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://cursor.com/docs/agent/overview Cursor handles this as a draft: Use the hours I said I have. Do not move my shifts. ## Prompt You are working as Cursor on one playbook. Task: Use the hours I said I have. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not move my shifts. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Cursor where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Use the hours I said I have. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not move my shifts. # Plan a grocery list from a week's dinners URL: https://aiplaybookgems.com/usecases/openclaw-plan-a-grocery-list-from-a-week-s-dinners Agent: OpenClaw Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://openclaw.ai/ OpenClaw handles this as a draft: Deduplicate ingredients from the meals I name. Mark staples I said I already have. Do not order delivery. ## Prompt You are working as OpenClaw on one playbook. Task: Deduplicate ingredients from the meals I name. Mark staples I said I already have. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not order delivery. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open OpenClaw where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Deduplicate ingredients from the meals I name. Mark staples I said I already have. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not order delivery. # Prep a pre-op checklist from the hospital packet URL: https://aiplaybookgems.com/usecases/claude-prep-a-pre-op-checklist-from-the-hospital-packet Agent: Claude Category: Operations Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Stop-food time, what to bring, and the phone number on the packet. Do not call the hospital. ## Prompt You are working as Claude on one playbook. Task: Stop-food time, what to bring, and the phone number on the packet. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not call the hospital. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Stop-food time, what to bring, and the phone number on the packet. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not call the hospital. # Propose a weekly ops checklist from last month's misses URL: https://aiplaybookgems.com/usecases/grok-bot-propose-a-weekly-ops-checklist-from-last-month-s-misses Agent: Grok Bot Category: Operations Roles: Operator Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: Use the incident notes I paste. Five checks, each tied to a miss. Do not change production. ## Prompt You are working as Grok Bot on one playbook. Task: Use the incident notes I paste. Five checks, each tied to a miss. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not change production. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Use the incident notes I paste. Five checks, each tied to a miss. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not change production. # Pull exercise notes into a weekly view URL: https://aiplaybookgems.com/usecases/hermes-agent-pull-exercise-notes-into-a-weekly-view Agent: Hermes Agent Category: Operations Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: What I logged. No new plan. Do not invent workouts. ## Prompt You are working as Hermes Agent on one playbook. Task: What I logged. No new plan. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not invent workouts. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What I logged. No new plan. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not invent workouts. # Sort allowance and chore notes URL: https://aiplaybookgems.com/usecases/hermes-agent-sort-allowance-and-chore-notes Agent: Hermes Agent Category: Operations Roles: Parent Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Child, chore, and the amount I already decided. Do not invent amounts. Do not transfer money. ## Prompt You are working as Hermes Agent on one playbook. Task: Child, chore, and the amount I already decided. Do not invent amounts. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not transfer money. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Child, chore, and the amount I already decided. Do not invent amounts. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not transfer money. # Storyboard a 30-second demo URL: https://aiplaybookgems.com/usecases/crewai-storyboard-a-30-second-demo Agent: CrewAI Category: Sales Roles: Designer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.crewai.com/introduction CrewAI handles this as a draft: Six frames and the UI I said is real. Do not record the demo. ## Prompt You are working as CrewAI on one playbook. Task: Six frames and the UI I said is real. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not record the demo. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open CrewAI where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Six frames and the UI I said is real. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not record the demo. # Summarize a angry thread for a manager URL: https://aiplaybookgems.com/usecases/n8n-summarize-a-angry-thread-for-a-manager Agent: n8n Category: Support Roles: Support lead Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://docs.n8n.io/build/build-and-manage-agents n8n handles this as a draft: What they asked, what we did, and the open promise. Do not reply in the thread. ## Prompt You are working as n8n on one playbook. Task: What they asked, what we did, and the open promise. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not reply in the thread. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open n8n where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What they asked, what we did, and the open promise. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not reply in the thread. # Summarize reviews for one failure mode URL: https://aiplaybookgems.com/usecases/perplexity-summarize-reviews-for-one-failure-mode Agent: Perplexity Category: Research Roles: Operator Industries: Retail Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: From the reviews I paste, only comments about the failure I name. Count them. Do not scrape a site I did not paste. ## Prompt You are working as Perplexity on one playbook. Task: From the reviews I paste, only comments about the failure I name. Count them. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not scrape a site I did not paste. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the reviews I paste, only comments about the failure I name. Count them. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not scrape a site I did not paste. # Turn a call recording note into a case comment URL: https://aiplaybookgems.com/usecases/zapier-agents-turn-a-call-recording-note-into-a-case-comment Agent: Zapier Agents Category: Support Roles: Support lead Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://zapier.com/agents Zapier Agents handles this as a draft: Facts only, under 150 words. Do not email the customer. ## Prompt You are working as Zapier Agents on one playbook. Task: Facts only, under 150 words. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not email the customer. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Zapier Agents where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Facts only, under 150 words. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not email the customer. # Turn a dataset dictionary into questions URL: https://aiplaybookgems.com/usecases/perplexity-turn-a-dataset-dictionary-into-questions Agent: Perplexity Category: Data Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Which fields are missing descriptions. Do not query production data. ## Prompt You are working as Perplexity on one playbook. Task: Which fields are missing descriptions. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not query production data. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Which fields are missing descriptions. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not query production data. # Turn a literature table into gaps URL: https://aiplaybookgems.com/usecases/chatgpt-agent-turn-a-literature-table-into-gaps Agent: ChatGPT Agent Category: Research Roles: Student Industries: Education Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://developers.openai.com/api/docs/guides/tools-computer-use ChatGPT Agent handles this as a draft: What the rows do not cover, in one list. Do not claim the gap is unstudied beyond this table. ## Prompt You are working as ChatGPT Agent on one playbook. Task: What the rows do not cover, in one list. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not claim the gap is unstudied beyond this table. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open ChatGPT Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What the rows do not cover, in one list. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not claim the gap is unstudied beyond this table. # Turn a problem set into a hint ladder URL: https://aiplaybookgems.com/usecases/claude-turn-a-problem-set-into-a-hint-ladder Agent: Claude Category: Research Roles: Student Industries: Education Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Hints that do not reveal the final answer. I still solve it. ## Prompt You are working as Claude on one playbook. Task: Hints that do not reveal the final answer. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: I still solve it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Hints that do not reveal the final answer. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: I still solve it. # Turn voice memos into a household decision list URL: https://aiplaybookgems.com/usecases/lindy-turn-voice-memos-into-a-household-decision-list Agent: Lindy Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: Transcribe the decisions, the options, and who I said should weigh in. Do not message those people. ## Prompt You are working as Lindy on one playbook. Task: Transcribe the decisions, the options, and who I said should weigh in. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not message those people. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Transcribe the decisions, the options, and who I said should weigh in. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not message those people. # Write a one-page net-worth snapshot URL: https://aiplaybookgems.com/usecases/claude-cowork-write-a-one-page-net-worth-snapshot Agent: Claude Cowork Category: Data Roles: Founder Industries: Finance Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Assets and debts I paste, with a note when a number is an estimate I marked as such. Do not pull accounts on your own. ## Prompt You are working as Claude Cowork on one playbook. Task: Assets and debts I paste, with a note when a number is an estimate I marked as such. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not pull accounts on your own. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Assets and debts I paste, with a note when a number is an estimate I marked as such. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not pull accounts on your own. # A two-role CrewAI crew that argues with its own draft URL: https://aiplaybookgems.com/usecases/crewai-research-crew Agent: CrewAI Category: Research Roles: Researcher Industries: Software Difficulty: hard Cost bucket: free Estimated cost: The library is open source. You pay the model provider. No CrewAI cloud plan is required for this playbook. Time saved: Little on the first setup. More once the same crew runs on a series of questions. Status: unverified Source: https://docs.crewai.com/introduction A researcher agent gathers sources and a reviewer agent rejects unsourced lines. You run it locally from the CrewAI docs' model. ## Prompt Researcher role: gather notes on the QUESTION using tools the project already configured. Every note needs a URL. No URL, no note. Reviewer role: delete any sentence without a URL. Return the surviving brief plus a list of deleted claims. Do not add new facts during review. QUESTION: ## Steps 1. Install CrewAI from the current introduction: docs.crewai.com/introduction is the index. Pin the version you install in the project. 2. Define two agents and one task each: Researcher then reviewer. Give the reviewer no search tool so it cannot sneak in new facts. 3. Put the prompt text in the task descriptions: Run a question you already know. The deleted-claims list should be non-empty at least once, or the reviewer is rubber-stamping. 4. Read the surviving URLs: A crew that cites homepages for specific numbers is failing the spirit of the prompt. # Answer questions from a Google Sheet without editing it URL: https://aiplaybookgems.com/usecases/n8n-sheets-answers Agent: n8n Category: Data Roles: Operations Industries: Ecommerce Difficulty: medium Cost bucket: low Estimated cost: n8n plus a model. The sheet is whatever you already use. Time saved: Repeated lookups across a catalog or policy sheet. Status: unverified Source: https://docs.n8n.io/build/build-and-manage-agents An n8n agent uses a sheet as its knowledge source and answers in a private chat. The sheet stays read-only. ## Prompt Answer questions using only the connected Google Sheet. If the sheet does not contain the answer, say "not in the sheet" and name the columns you checked. Quote the row's key column in every answer. Do not update, append, or format the sheet. Do not use outside knowledge for prices, stock, or policy. If two rows conflict, show both and do not pick a winner. ## Steps 1. Clean the sheet so a column means one thing: Merged cells and notes in the margin become fake answers. Fix that before connecting an agent. 2. Create an n8n agent and attach the sheet as a tool or knowledge source: The agent docs explicitly mention Google Sheets among the tools an agent can use. 3. Ask five questions you know the answers to: Include one trick question that is not in the sheet. The correct reply is 'not in the sheet'. 4. Leave the credential read-only: The prompt asks for no writes. The credential should make that true even if the prompt is edited later. # Remind me to check in, then stop for a yes URL: https://aiplaybookgems.com/usecases/openclaw-flight-checkin Agent: OpenClaw Category: Operations Roles: Operations Industries: Logistics Difficulty: easy Cost bucket: free Estimated cost: Software is free to run yourself. Airline accounts and the model are your own. Time saved: Missing a check-in window is the expensive failure. The reminder itself is a minute. Status: unverified Source: https://openclaw.ai/ OpenClaw's site says it can check you in for flights from chat. This playbook keeps the reminder, and requires a yes before any airline page is submitted. ## Prompt Watch the trip I paste below. 24 hours before departure, message me in this chat: airline, route, departure local time, and the check-in link if it is in the confirmation. Do not open a payment page. Do not submit check-in, choose seats, or accept fees unless I reply "check me in". If the confirmation has no record locator, ask me for it. Do not guess. When I say "check me in", tell me which page you are on before you submit any form, and stop if the airline asks for payment. TRIP CONFIRMATION: ## Steps 1. Paste a confirmation into the chat you already use: The OpenClaw homepage lists flight check-in alongside inbox and calendar. Keep the record locator out of group chats. 2. Ask for a reminder on a past trip first: Use an already-flown reservation to see the message shape without a live check-in window. 3. Require the phrase before submit: The prompt stops on payment pages even after you agree. 4. Prefer the airline's official app if the bot gets lost: A missed button is annoying. A wrong paid seat is not. # Ask Copilot for a test plan, then run it yourself URL: https://aiplaybookgems.com/usecases/copilot-test-plan Agent: GitHub Copilot Category: Coding Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: GitHub Copilot subscription. Time saved: Remembering edge cases, not executing them. Status: unverified Source: https://docs.github.com/en/copilot/get-started/what-is-github-copilot Copilot lists cases for a change you describe. You run them. The plan is not permission to skip the ones that are slow. ## Prompt Here is the change. Write a test plan I can execute. Include: the happy path, two boundary cases, one failure case, and one case that should stay unchanged. For each case give the setup, the action, and the observable result. Prefer a command or a user step that exists today. Mark cases that need a staging account as manual. Do not claim you ran them. I will run them. CHANGE: ## Steps 1. Describe the change in behavior, not file names only: Copilot can see the IDE. Still write the user-visible change so the plan is about behavior. 2. Generate the plan in Copilot chat: Use the IDE chat documented in GitHub's Copilot getting-started materials. 3. Run the cases and check them off: Paste failures back if you want a fix. The plan itself is not the fix. 4. Put the plan in the PR: A reviewer can see what you actually did versus what was listed. # Collect a competitor pack into one folder of files URL: https://aiplaybookgems.com/usecases/manus-competitor-pack Agent: Manus Category: Research Roles: Marketer Industries: Media Difficulty: hard Cost bucket: medium Estimated cost: Manus usage for a long browsing task. Treat a full pack as a larger run than a single memo. Time saved: The collection pass. Analysis still takes a person. Status: unverified Source: https://manus.im/docs/introduction/welcome Manus gathers public pages, pricing screenshots, and a contradiction list into files you can drop into a doc. It does not publish anything. ## Prompt Build a competitor pack for the company I name and two peers I name. For each, save: - A text extract of the homepage promise, with the URL - The pricing page URL and the plan names visible without logging in - One screenshot of pricing if the page is public - A list of claims that contradict each other across those pages Do not sign up, do not post, do not contact them. Put the files in one folder and include a README that lists what you could not access. COMPANY: PEERS: ## Steps 1. Name the company and two peers: More than three sites turns the check into a weekend. The Manus API docs are there if you later want this from your own app. This playbook uses the product, not the API. 2. Ask for a folder, not a chat summary: The welcome page frames Manus as delivering a result. A folder is reviewable. A monologue is not. 3. Spot-check screenshots against the live page: Pricing pages change. The README of failures matters as much as the captures. 4. Do not ship the pack externally the same day: A contradiction list can be wrong because a page was mid-deploy. Someone on the team should look. # Map a repo for someone who cloned it this morning URL: https://aiplaybookgems.com/usecases/claude-code-repo-map Agent: Claude Code Category: Coding Roles: Software engineer, Product manager Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One Claude Code session. No extra services. Time saved: A half day of wandering for a new teammate, if the map is checked by someone who already knows the repo. Status: unverified Source: https://code.claude.com/docs/en/overview Claude Code writes a one-page map of how to run, test, and change the system. It does not edit application code. ## Prompt Do not edit application code. You may add a single doc if I ask for a file at the end. Read the repository and write a map with: - How to install and run tests, using scripts that actually exist. Quote the script names. - The five directories a newcomer will touch first, and what lives there. - Where configuration and secrets are expected. Name the env example file if there is one. Do not print secret values. - One golden path: the function or route that handles the core action, with file paths. - Known traps you can see in comments, CI config, or docs. If you are unsure, say unsure. Do not invent a makefile. ## Steps 1. Run Claude Code from a clean clone: The overview docs cover opening a project. A clean clone shows missing setup the old checkout hides. 2. Paste the prompt and forbid code edits: The first sentence is the safety rail. Stop the session if files start changing. 3. Have a current owner correct the map: A confident wrong golden path is worse than no map. Mark corrections in the doc. 4. Commit the doc only after that read: Ask for the file in a second turn so the draft can be fixed first. # Plan the week from the calendar without moving it URL: https://aiplaybookgems.com/usecases/lindy-plan-the-week Agent: Lindy Category: Operations Roles: Founder Industries: Agencies Difficulty: easy Cost bucket: medium Estimated cost: Lindy subscription, plus the calendar you already use. Time saved: The Sunday sorting of a crowded week, maybe an hour, if the calendar is the real source of truth. Status: unverified Source: https://lindy.ai/ Lindy's site lists 'Plan your week'. This playbook returns a weekly plan and proposed moves. The calendar stays put until you edit it. ## Prompt Read next week's calendar. Produce a one-page plan: - The three outcomes that have time blocked, and the three that do not - Meetings that could move, with the reason - A proposed day shape for the one day that is most fragmented Do not send email, decline invites, or edit the calendar. I will move things myself. Treat holds named "focus" or "no meetings" as busy. ## Steps 1. Connect the calendar Lindy should read: Use the account where the real week lives. The site lists weekly planning as a use case. 2. Run the prompt on a week you remember: If it proposes moving a meeting that already happened, the timezone or calendar is wrong. 3. Apply moves yourself: The value is the list of conflicts, not an agent rearranging other people's days. 4. Keep a focus block in the test week: The last line of the prompt exists so empty-looking holds are not treated as free. # Build a carry-on kit for a red-eye with a child URL: https://aiplaybookgems.com/usecases/perplexity-build-a-carry-on-kit-for-a-red-eye-with-a-child Agent: Perplexity Category: Operations Roles: Parent Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Items from my house list that fit a personal item. Do not shop. ## Prompt You are working as Perplexity on one playbook. Task: Items from my house list that fit a personal item. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not shop. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Items from my house list that fit a personal item. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not shop. # Compare two savings yields I paste URL: https://aiplaybookgems.com/usecases/claude-compare-two-savings-yields-i-paste Agent: Claude Category: Research Roles: Operator Industries: Finance Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Table the rate, the cap, and the fee. Say which number you could not find on the page. Do not open an account. ## Prompt You are working as Claude on one playbook. Task: Table the rate, the cap, and the fee. Say which number you could not find on the page. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not open an account. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Table the rate, the cap, and the fee. Say which number you could not find on the page. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not open an account. # Draft a bug report from a customer description URL: https://aiplaybookgems.com/usecases/lindy-draft-a-bug-report-from-a-customer-description Agent: Lindy Category: Support Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: Steps, expected, actual. Mark unknowns. Do not file it if the tracker needs an engineer. ## Prompt You are working as Lindy on one playbook. Task: Steps, expected, actual. Mark unknowns. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not file it if the tracker needs an engineer. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Steps, expected, actual. Mark unknowns. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not file it if the tracker needs an engineer. # Draft a discussion post from my notes URL: https://aiplaybookgems.com/usecases/hermes-agent-draft-a-discussion-post-from-my-notes Agent: Hermes Agent Category: Research Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: My point and one quote from the reading. Do not post it. ## Prompt You are working as Hermes Agent on one playbook. Task: My point and one quote from the reading. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not post it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: My point and one quote from the reading. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not post it. # Draft a marketplace listing for something I own URL: https://aiplaybookgems.com/usecases/claude-draft-a-marketplace-listing-for-something-i-own Agent: Claude Category: Operations Roles: Operator Industries: Retail Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Title, condition from my words, and flaws I listed. No invented features. Do not post the listing. ## Prompt You are working as Claude on one playbook. Task: Title, condition from my words, and flaws I listed. No invented features. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not post the listing. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Title, condition from my words, and flaws I listed. No invented features. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not post the listing. # Draft a newsletter intro from three links URL: https://aiplaybookgems.com/usecases/gemini-draft-a-newsletter-intro-from-three-links Agent: Gemini Category: Sales Roles: Marketer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: One sentence per link using the title I pasted. Do not send the newsletter. ## Prompt You are working as Gemini on one playbook. Task: One sentence per link using the title I pasted. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send the newsletter. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: One sentence per link using the title I pasted. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send the newsletter. # Draft a teacher absence email URL: https://aiplaybookgems.com/usecases/claude-draft-a-teacher-absence-email Agent: Claude Category: Operations Roles: Parent Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Dates and the work I want sent home, in my voice, under 100 words. Do not send it. ## Prompt You are working as Claude on one playbook. Task: Dates and the work I want sent home, in my voice, under 100 words. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Dates and the work I want sent home, in my voice, under 100 words. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send it. # Draft show notes from a timestamp list URL: https://aiplaybookgems.com/usecases/claude-cowork-draft-show-notes-from-a-timestamp-list Agent: Claude Cowork Category: Sales Roles: Marketer Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Topic at each timestamp I marked. Do not publish. ## Prompt You are working as Claude Cowork on one playbook. Task: Topic at each timestamp I marked. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not publish. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Topic at each timestamp I marked. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not publish. # Extract definitions of a metric across docs URL: https://aiplaybookgems.com/usecases/crewai-extract-definitions-of-a-metric-across-docs Agent: CrewAI Category: Research Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Forty minutes Status: unverified Source: https://docs.crewai.com/introduction CrewAI handles this as a draft: Each doc's definition, quoted. Do not pick one as official. ## Prompt You are working as CrewAI on one playbook. Task: Each doc's definition, quoted. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not pick one as official. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open CrewAI where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Each doc's definition, quoted. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not pick one as official. # Extract sizes from a clothing page for a gift URL: https://aiplaybookgems.com/usecases/gemini-extract-sizes-from-a-clothing-page-for-a-gift Agent: Gemini Category: Research Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: The size chart numbers and the one question I should ask the recipient. Do not order. ## Prompt You are working as Gemini on one playbook. Task: The size chart numbers and the one question I should ask the recipient. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not order. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: The size chart numbers and the one question I should ask the recipient. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not order. # List insurance questions before a procedure URL: https://aiplaybookgems.com/usecases/gemini-list-insurance-questions-before-a-procedure Agent: Gemini Category: Data Roles: Operator Industries: Healthcare Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: From the estimate: what is covered and what the estimate says I might owe. Do not schedule the procedure. ## Prompt You are working as Gemini on one playbook. Task: From the estimate: what is covered and what the estimate says I might owe. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not schedule the procedure. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the estimate: what is covered and what the estimate says I might owe. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not schedule the procedure. # List macros that contradict the new policy URL: https://aiplaybookgems.com/usecases/claude-list-macros-that-contradict-the-new-policy Agent: Claude Category: Support Roles: Support lead Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Forty minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Quote both lines. Do not edit the macros. ## Prompt You are working as Claude on one playbook. Task: Quote both lines. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not edit the macros. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Quote both lines. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not edit the macros. # List objections from lost-deal notes URL: https://aiplaybookgems.com/usecases/zapier-agents-list-objections-from-lost-deal-notes Agent: Zapier Agents Category: Sales Roles: Founder Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Forty minutes Status: unverified Source: https://zapier.com/agents Zapier Agents handles this as a draft: The objection in their words and how often it appears. Do not write a battlecard yet. ## Prompt You are working as Zapier Agents on one playbook. Task: The objection in their words and how often it appears. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not write a battlecard yet. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Zapier Agents where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: The objection in their words and how often it appears. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not write a battlecard yet. # List unclaimed-property search results worth opening URL: https://aiplaybookgems.com/usecases/hermes-agent-list-unclaimed-property-search-results-worth-opening Agent: Hermes Agent Category: Research Roles: Operator Industries: Finance Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: A careful half hour Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: From the search results I paste, keep only rows that match my name and a state I have lived in. Say what page I should open next. Do not submit a claim. ## Prompt You are working as Hermes Agent on one playbook. Task: From the search results I paste, keep only rows that match my name and a state I have lived in. Say what page I should open next. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not submit a claim. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the search results I paste, keep only rows that match my name and a state I have lived in. Say what page I should open next. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not submit a claim. # Plan a holiday travel roster URL: https://aiplaybookgems.com/usecases/claude-cowork-plan-a-holiday-travel-roster Agent: Claude Cowork Category: Operations Roles: Parent Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Who is coming, which nights, and the beds I said I have. Flag overbooking. Do not email relatives. ## Prompt You are working as Claude Cowork on one playbook. Task: Who is coming, which nights, and the beds I said I have. Flag overbooking. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not email relatives. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Who is coming, which nights, and the beds I said I have. Flag overbooking. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not email relatives. # Plan a rail day from timetables I paste URL: https://aiplaybookgems.com/usecases/chatgpt-agent-plan-a-rail-day-from-timetables-i-paste Agent: ChatGPT Agent Category: Research Roles: Operator Industries: Hospitality Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://developers.openai.com/api/docs/guides/tools-computer-use ChatGPT Agent handles this as a draft: Connections with at least the buffer I set. Flag tight transfers. Do not buy tickets. ## Prompt You are working as ChatGPT Agent on one playbook. Task: Connections with at least the buffer I set. Flag tight transfers. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not buy tickets. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open ChatGPT Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Connections with at least the buffer I set. Flag tight transfers. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not buy tickets. # Prepare a move inventory from room photos URL: https://aiplaybookgems.com/usecases/grok-bot-prepare-a-move-inventory-from-room-photos Agent: Grok Bot Category: Operations Roles: Operator Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An afternoon Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: Room, item, and condition in one line each. Say when a photo is too dark. Do not post the list. ## Prompt You are working as Grok Bot on one playbook. Task: Room, item, and condition in one line each. Say when a photo is too dark. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not post the list. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Room, item, and condition in one line each. Say when a photo is too dark. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not post the list. # Reconcile a partner's shared travel fund URL: https://aiplaybookgems.com/usecases/gemini-reconcile-a-partner-s-shared-travel-fund Agent: Gemini Category: Data Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Match transfers to trip costs I tag. Show the balance. Do not move the fund. ## Prompt You are working as Gemini on one playbook. Task: Match transfers to trip costs I tag. Show the balance. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not move the fund. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Match transfers to trip costs I tag. Show the balance. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not move the fund. # Sort medical PDFs by visit date URL: https://aiplaybookgems.com/usecases/perplexity-sort-medical-pdfs-by-visit-date Agent: Perplexity Category: Data Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Propose filenames. Do not delete. Leave the files in place. ## Prompt You are working as Perplexity on one playbook. Task: Propose filenames. Do not delete. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Leave the files in place. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Propose filenames. Do not delete. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Leave the files in place. # Spot duplicate pharmacy charges URL: https://aiplaybookgems.com/usecases/gemini-spot-duplicate-pharmacy-charges Agent: Gemini Category: Data Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Under an hour Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Group charges by pharmacy and date. Show pairs that look like the same fill billed twice. Do not call the pharmacy. ## Prompt You are working as Gemini on one playbook. Task: Group charges by pharmacy and date. Show pairs that look like the same fill billed twice. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not call the pharmacy. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Group charges by pharmacy and date. Show pairs that look like the same fill billed twice. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not call the pharmacy. # Summarize a homeowners association packet URL: https://aiplaybookgems.com/usecases/hermes-agent-summarize-a-homeowners-association-packet Agent: Hermes Agent Category: Research Roles: Operator Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Forty minutes Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: List votes, dues changes, and deadlines. Quote the page or section. Do not vote or pay. ## Prompt You are working as Hermes Agent on one playbook. Task: List votes, dues changes, and deadlines. Quote the page or section. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not vote or pay. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: List votes, dues changes, and deadlines. Quote the page or section. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not vote or pay. # Summarize a paper's method in one page URL: https://aiplaybookgems.com/usecases/gemini-summarize-a-paper-s-method-in-one-page Agent: Gemini Category: Research Roles: Student Industries: Education Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Data, method, and the limit the authors name. Do not add results that are not in the PDF. ## Prompt You are working as Gemini on one playbook. Task: Data, method, and the limit the authors name. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not add results that are not in the PDF. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Data, method, and the limit the authors name. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not add results that are not in the PDF. # Summarize a policy change for the team URL: https://aiplaybookgems.com/usecases/n8n-summarize-a-policy-change-for-the-team Agent: n8n Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Twenty minutes Status: unverified Source: https://docs.n8n.io/build/build-and-manage-agents n8n handles this as a draft: What changed, who it affects, and the effective date. Quote the memo. Do not announce it. ## Prompt You are working as n8n on one playbook. Task: What changed, who it affects, and the effective date. Quote the memo. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not announce it. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open n8n where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What changed, who it affects, and the effective date. Quote the memo. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not announce it. # Summarize a regulatory page I pasted URL: https://aiplaybookgems.com/usecases/gemini-summarize-a-regulatory-page-i-pasted Agent: Gemini Category: Research Roles: Founder Industries: Finance Difficulty: medium Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Who it applies to and the effective date if printed. Do not give legal advice. ## Prompt You are working as Gemini on one playbook. Task: Who it applies to and the effective date if printed. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not give legal advice. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Who it applies to and the effective date if printed. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not give legal advice. # Turn a CSV of office requests into themes URL: https://aiplaybookgems.com/usecases/lindy-turn-a-csv-of-office-requests-into-themes Agent: Lindy Category: Data Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://lindy.ai/ Lindy handles this as a draft: Top five request types and an example row for each. Do not fulfill a request. ## Prompt You are working as Lindy on one playbook. Task: Top five request types and an example row for each. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not fulfill a request. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Lindy where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Top five request types and an example row for each. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not fulfill a request. # Turn a stack trace into a suspect list URL: https://aiplaybookgems.com/usecases/github-copilot-turn-a-stack-trace-into-a-suspect-list Agent: GitHub Copilot Category: Coding Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://docs.github.com/en/copilot/get-started/what-is-github-copilot GitHub Copilot handles this as a draft: Frames in our code, not the library, first. Do not push a fix. ## Prompt You are working as GitHub Copilot on one playbook. Task: Frames in our code, not the library, first. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not push a fix. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open GitHub Copilot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Frames in our code, not the library, first. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not push a fix. # Turn a ticket into an implementation note URL: https://aiplaybookgems.com/usecases/cursor-turn-a-ticket-into-an-implementation-note Agent: Cursor Category: Coding Roles: Software engineer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://cursor.com/docs/agent/overview Cursor handles this as a draft: Acceptance checks from the ticket text. Do not start coding until I say so. ## Prompt You are working as Cursor on one playbook. Task: Acceptance checks from the ticket text. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not start coding until I say so. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Cursor where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Acceptance checks from the ticket text. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not start coding until I say so. # Write alt text and a caption for a screenshot URL: https://aiplaybookgems.com/usecases/manus-write-alt-text-and-a-caption-for-a-screenshot Agent: Manus Category: Sales Roles: Marketer Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Describe what is visible. No claims about results. Do not publish. ## Prompt You are working as Manus on one playbook. Task: Describe what is visible. No claims about results. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not publish. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Describe what is visible. No claims about results. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not publish. # Fix keyboard and name issues on one screen URL: https://aiplaybookgems.com/usecases/cursor-accessibility-pass Agent: Cursor Category: Coding Roles: Software engineer, Designer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: Cursor subscription. A browser is already on the machine. Time saved: The first pass of an accessibility punch list on a single screen. Status: unverified Source: https://cursor.com/docs/agent/overview Cursor Agent walks one screen for missing names, keyboard traps, and contrast notes, then patches the clear code issues. ## Prompt Look only at the screen reached by the route I name. Do not restyle the app. Find: - Buttons or links with no accessible name - Inputs with no label - Click handlers on elements that are not keyboard reachable - Images of text or icons that convey meaning without a text alternative Fix the ones that are clearly code bugs. For contrast and focus order, write notes instead of guessing a new visual design. List each change with the file and the user-facing result. Do not claim WCAG conformance. ROUTE: ## Steps 1. Pick one route: A marketing site plus the app is not one pass. Name the path. 2. Run Agent in the repo: Ask it to search for the route's components before editing. 3. Tab through the screen yourself: The agent will miss things a keyboard pass catches, and it may 'fix' a control your design system already names. 4. File the contrast notes separately: The prompt keeps visual redesign out of the diff. # Route a support ticket in LangGraph and pause before reply URL: https://aiplaybookgems.com/usecases/langgraph-ticket-router Agent: LangGraph Category: Support Roles: Support lead Industries: Software Difficulty: hard Cost bucket: free Estimated cost: Open-source library. Model calls are the cost. LangSmith is optional and not required here. Time saved: Sorting a repetitive queue, after the graph is built. Building it is real engineering. Status: unverified Source: https://docs.langchain.com/oss/javascript/langgraph/overview A small LangGraph app classifies a ticket, drafts a reply, and interrupts for a person before anything is sent. ## Prompt Classify the ticket as billing, bug, or how-to. Draft a reply using only the help-center excerpts in the prompt. If the excerpts do not cover it, draft a handoff note to a human instead of an answer. Then interrupt and wait. Do not call a send tool. The graph may continue only when the state field approved is true. TICKET: EXCERPTS: ## Steps 1. Start from the LangGraph overview and quickstart: The docs cover persistence, interrupts, and human-in-the-loop. Use an interrupt for the approval rather than a comment in the prompt alone. 2. Model three nodes: classify, draft, wait: The wait node does not have a send tool. That is the product decision. 3. Feed it ten old tickets with known labels: Write down disagreements. Fix the excerpts before you tune adjectives in the prompt. 4. Only then consider a send tool: Add it behind the approved flag, in a separate change, with a log of who approved. # Turn a Slack report into a Linear issue draft URL: https://aiplaybookgems.com/usecases/n8n-linear-drafts Agent: n8n Category: Coding Roles: Product manager Industries: Software Difficulty: medium Cost bucket: low Estimated cost: n8n and the model. Linear is your existing workspace. Time saved: Rewriting messy reports into the same issue template. Status: unverified Source: https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.agent The n8n agent drafts a Linear issue from a bug report in Slack and shows it to you before anything is filed. ## Prompt When I forward a Slack message, draft a Linear issue with: title, environment, steps, expected, actual, and a link back to the Slack message. Do not file the issue, assign it, or change priority until I reply APPROVE and a team key. If steps are missing, put "not provided" in that section. Do not invent a reproduction. Keep the title under 80 characters and free of adjectives. ## Steps 1. Connect Slack and Linear: n8n's agent documentation names both Slack and Linear as action tools. Start with a personal Linear key that cannot edit production projects if you have that split. 2. Use the prompt as the agent instructions: Forward one real report. Check that missing steps stay missing. 3. Approve with the team key in a second message: Filing is a separate action so a bad draft never lands in the tracker. 4. Compare the filed issue with the draft: If the agent adds scope during filing, narrow the tool to a single create action. # Implement a small feature that already has acceptance notes URL: https://aiplaybookgems.com/usecases/codex-cli-small-feature Agent: Codex CLI Category: Coding Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: Codex on a ChatGPT plan. Local compute is yours. Time saved: A focused implementation pass when acceptance notes already exist. Status: unverified Source: https://learn.chatgpt.com/docs/codex/cli Codex CLI builds a feature described as observable behavior, runs the tests you name, and leaves unrelated files alone. ## Prompt Implement the feature described in ACCEPTANCE. Those lines are the spec. Do not add settings, flags, or screens they do not mention. Follow AGENTS.md if the repo has one. If it conflicts with ACCEPTANCE, stop and ask. Run the test command. Update or add tests so each acceptance line has one assertion you can point to. Do not commit unless I ask. Show the diff stat first. ACCEPTANCE: TEST COMMAND: ## Steps 1. Write acceptance as observable behavior: 'Make it faster' is not a line Codex can satisfy honestly. 'The export includes a created_at column' is. 2. Start Codex CLI in the repo: The CLI docs are the install source. If the repo has AGENTS.md, the prompt tells Codex to follow it. 3. Check that each acceptance line has a test: The prompt requires a pointer from line to assertion. Missing pointers go back into the session. 4. Commit when the diff stat matches the spec: You commit. The prompt refuses to commit on its own. # Review a Devin session before anyone merges URL: https://aiplaybookgems.com/usecases/devin-review-the-session Agent: Devin Category: Coding Roles: Engineering manager Industries: Software Difficulty: medium Cost bucket: medium Estimated cost: The Devin plan you already used to produce the session. Time saved: Reconstructing what the agent did, which is often longer than reading a normal PR. Status: unverified Source: https://docs.devin.ai/get-started/devin-intro Use Devin's session output as a review packet: commands run, files touched, and tests actually executed. The playbook is for the reviewer, not a second coding agent. ## Prompt I am reviewing your last session. Do not push more commits unless I say "amend". Answer in this order: - Commits or PRs you opened, with URLs - Commands you ran, and which ones failed - Files you edited that the issue did not mention - Tests you did not run - Secrets or production systems you touched If a point is unknown because the session log does not show it, say unknown. Do not reconstruct a flattering story. ## Steps 1. Open the finished Devin session: Use the product's session view from the intro docs' first-session flow. This prompt is a review, so start it after the PR exists. 2. Paste the prompt into that session: The ban on extra commits matters. A review question should not become another push. 3. Compare the command list with CI: If CI ran checks the session skipped, read those logs yourself. 4. Decide merge, changes, or close: The manager's job in this playbook is the decision, not another prompt. # Build a family seat map request URL: https://aiplaybookgems.com/usecases/gemini-build-a-family-seat-map-request Agent: Gemini Category: Operations Roles: Parent Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Who needs to sit together, from the passenger list I write. Do not change the booking. ## Prompt You are working as Gemini on one playbook. Task: Who needs to sit together, from the passenger list I write. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not change the booking. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Who needs to sit together, from the passenger list I write. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not change the booking. # Build a rent-increase reply from the lease URL: https://aiplaybookgems.com/usecases/grok-bot-build-a-rent-increase-reply-from-the-lease Agent: Grok Bot Category: Operations Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Thirty minutes with the lease Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: Quote the notice period and the increase. Draft a short reply that asks for the calculation. Leave blanks where the lease is silent. Do not send the reply. ## Prompt You are working as Grok Bot on one playbook. Task: Quote the notice period and the increase. Draft a short reply that asks for the calculation. Leave blanks where the lease is silent. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send the reply. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Quote the notice period and the increase. Draft a short reply that asks for the calculation. Leave blanks where the lease is silent. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send the reply. # Compare a refund request to the policy URL: https://aiplaybookgems.com/usecases/langgraph-compare-a-refund-request-to-the-policy Agent: LangGraph Category: Support Roles: Support lead Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://docs.langchain.com/oss/javascript/langgraph/overview LangGraph handles this as a draft: Approve-shaped or deny-shaped, with the policy sentence. I decide. Do not issue a refund. ## Prompt You are working as LangGraph on one playbook. Task: Approve-shaped or deny-shaped, with the policy sentence. I decide. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not issue a refund. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open LangGraph where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Approve-shaped or deny-shaped, with the policy sentence. I decide. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not issue a refund. # Compare shipping promises on two carts URL: https://aiplaybookgems.com/usecases/chatgpt-agent-compare-shipping-promises-on-two-carts Agent: ChatGPT Agent Category: Research Roles: Operator Industries: Ecommerce Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://developers.openai.com/api/docs/guides/tools-computer-use ChatGPT Agent handles this as a draft: Delivery window and shipping fee from the two pages. Nothing else. Do not place the order. ## Prompt You are working as ChatGPT Agent on one playbook. Task: Delivery window and shipping fee from the two pages. Nothing else. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not place the order. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open ChatGPT Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Delivery window and shipping fee from the two pages. Nothing else. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not place the order. # Draft a first response for a how-to gap URL: https://aiplaybookgems.com/usecases/n8n-draft-a-first-response-for-a-how-to-gap Agent: n8n Category: Support Roles: Support lead Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://docs.n8n.io/build/build-and-manage-agents n8n handles this as a draft: Link the closest article I name and say what it does not cover. Do not send. ## Prompt You are working as n8n on one playbook. Task: Link the closest article I name and say what it does not cover. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not send. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open n8n where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Link the closest article I name and say what it does not cover. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not send. # Draft a study-group agenda URL: https://aiplaybookgems.com/usecases/perplexity-draft-a-study-group-agenda Agent: Perplexity Category: Operations Roles: Student Industries: Education Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Three problems from the list I choose. Do not message the group. ## Prompt You are working as Perplexity on one playbook. Task: Three problems from the list I choose. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not message the group. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Three problems from the list I choose. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not message the group. # Draft a timeline for a launch checklist URL: https://aiplaybookgems.com/usecases/claude-cowork-draft-a-timeline-for-a-launch-checklist Agent: Claude Cowork Category: Operations Roles: Founder Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Dates I already chose, with the dependency I wrote. Do not announce the launch. ## Prompt You are working as Claude Cowork on one playbook. Task: Dates I already chose, with the dependency I wrote. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not announce the launch. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Dates I already chose, with the dependency I wrote. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not announce the launch. # Draft questions about a bill code URL: https://aiplaybookgems.com/usecases/claude-draft-questions-about-a-bill-code Agent: Claude Category: Data Roles: Operator Industries: Healthcare Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Fifteen minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: The code, the amount, and the question. No legal threat. Do not call. ## Prompt You are working as Claude on one playbook. Task: The code, the amount, and the question. No legal threat. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not call. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: The code, the amount, and the question. No legal threat. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not call. # File downloaded manuals by appliance URL: https://aiplaybookgems.com/usecases/claude-cowork-file-downloaded-manuals-by-appliance Agent: Claude Cowork Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Propose filenames and a folder for each PDF. Do not delete the originals. Do not throw files away. ## Prompt You are working as Claude Cowork on one playbook. Task: Propose filenames and a folder for each PDF. Do not delete the originals. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not throw files away. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Propose filenames and a folder for each PDF. Do not delete the originals. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not throw files away. # Find dead flags in a config file URL: https://aiplaybookgems.com/usecases/codex-cli-find-dead-flags-in-a-config-file Agent: Codex CLI Category: Coding Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://learn.chatgpt.com/docs/codex/cli Codex CLI handles this as a draft: Flags with no reference in the search I paste. Do not delete them. ## Prompt You are working as Codex CLI on one playbook. Task: Flags with no reference in the search I paste. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not delete them. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Codex CLI where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Flags with no reference in the search I paste. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not delete them. # List tax documents still missing URL: https://aiplaybookgems.com/usecases/claude-cowork-list-tax-documents-still-missing Agent: Claude Cowork Category: Data Roles: Finance Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Compare the checklist I write to the files in the folder. Return only the gaps. Do not file. ## Prompt You are working as Claude Cowork on one playbook. Task: Compare the checklist I write to the files in the folder. Return only the gaps. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not file. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Compare the checklist I write to the files in the folder. Return only the gaps. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not file. # Make a pantry restock from a photo of shelves URL: https://aiplaybookgems.com/usecases/manus-make-a-pantry-restock-from-a-photo-of-shelves Agent: Manus Category: Operations Roles: Operator Industries: Retail Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Fifteen minutes Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Items that look low. Say when the photo is unreadable. Do not place a grocery order. ## Prompt You are working as Manus on one playbook. Task: Items that look low. Say when the photo is unreadable. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not place a grocery order. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Items that look low. Say when the photo is unreadable. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not place a grocery order. # Make a reunion schedule from a shared doc URL: https://aiplaybookgems.com/usecases/openclaw-make-a-reunion-schedule-from-a-shared-doc Agent: OpenClaw Category: Operations Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://openclaw.ai/ OpenClaw handles this as a draft: Times, places, and who is bringing what. Flag blanks. Do not invite anyone. ## Prompt You are working as OpenClaw on one playbook. Task: Times, places, and who is bringing what. Flag blanks. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not invite anyone. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open OpenClaw where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Times, places, and who is bringing what. Flag blanks. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not invite anyone. # Make a source log for a memo URL: https://aiplaybookgems.com/usecases/manus-make-a-source-log-for-a-memo Agent: Manus Category: Research Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: URL, what I used it for, and the date on the page if present. Do not invent URLs. ## Prompt You are working as Manus on one playbook. Task: URL, what I used it for, and the date on the page if present. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not invent URLs. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: URL, what I used it for, and the date on the page if present. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not invent URLs. # Prep a one-page leave-behind URL: https://aiplaybookgems.com/usecases/perplexity-prep-a-one-page-leave-behind Agent: Perplexity Category: Sales Roles: Founder Industries: Agencies Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Problem, approach, and the proof point I already have. No new logos. Do not design or send. ## Prompt You are working as Perplexity on one playbook. Task: Problem, approach, and the proof point I already have. No new logos. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not design or send. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Problem, approach, and the proof point I already have. No new logos. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not design or send. # Propose a series of four posts from one essay URL: https://aiplaybookgems.com/usecases/claude-propose-a-series-of-four-posts-from-one-essay Agent: Claude Category: Sales Roles: Marketer Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: Each post one idea from the essay. Do not schedule them. ## Prompt You are working as Claude on one playbook. Task: Each post one idea from the essay. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not schedule them. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Each post one idea from the essay. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not schedule them. # Review dependency licenses I exported URL: https://aiplaybookgems.com/usecases/claude-code-review-dependency-licenses-i-exported Agent: Claude Code Category: Coding Roles: Software engineer Industries: Software Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://code.claude.com/docs/en/overview Claude Code handles this as a draft: Which ones are not on the allow list I wrote. Do not add packages. ## Prompt You are working as Claude Code on one playbook. Task: Which ones are not on the allow list I wrote. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not add packages. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Code where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Which ones are not on the allow list I wrote. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not add packages. # Summarize a municipal agenda URL: https://aiplaybookgems.com/usecases/perplexity-summarize-a-municipal-agenda Agent: Perplexity Category: Research Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Half an hour Status: unverified Source: https://docs.perplexity.ai/docs/getting-started/overview Perplexity handles this as a draft: Items and the action requested. Do not submit a public comment. ## Prompt You are working as Perplexity on one playbook. Task: Items and the action requested. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not submit a public comment. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Perplexity where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Items and the action requested. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not submit a public comment. # Summarize a RFP into a go or no-go brief URL: https://aiplaybookgems.com/usecases/grok-bot-summarize-a-rfp-into-a-go-or-no-go-brief Agent: Grok Bot Category: Sales Roles: Founder Industries: Software Difficulty: hard Cost bucket: medium Estimated cost: Expect a team seat, an automation platform, or steady model usage. Time saved: An hour Status: unverified Source: https://docs.x.ai/grok-bot/use-cases Grok Bot handles this as a draft: Requirements we match from my notes, and gaps. Do not submit the RFP. ## Prompt You are working as Grok Bot on one playbook. Task: Requirements we match from my notes, and gaps. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not submit the RFP. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Grok Bot where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Requirements we match from my notes, and gaps. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not submit the RFP. # Summarize a tutorial I finished URL: https://aiplaybookgems.com/usecases/claude-summarize-a-tutorial-i-finished Agent: Claude Category: Research Roles: Student Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://www.anthropic.com/news/3-5-models-and-computer-use Claude handles this as a draft: What I can now do, in my words from the checklist. Do not claim a certificate. ## Prompt You are working as Claude on one playbook. Task: What I can now do, in my words from the checklist. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not claim a certificate. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What I can now do, in my words from the checklist. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not claim a certificate. # Summarize a utility outage thread URL: https://aiplaybookgems.com/usecases/openclaw-summarize-a-utility-outage-thread Agent: OpenClaw Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://openclaw.ai/ OpenClaw handles this as a draft: What is out, the last update time, and the next promised update. Do not post in the thread. ## Prompt You are working as OpenClaw on one playbook. Task: What is out, the last update time, and the next promised update. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not post in the thread. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open OpenClaw where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: What is out, the last update time, and the next promised update. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not post in the thread. # Summarize a warranty claim I have not sent URL: https://aiplaybookgems.com/usecases/claude-cowork-summarize-a-warranty-claim-i-have-not-sent Agent: Claude Cowork Category: Operations Roles: Operator Industries: Retail Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Thirty minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Turn my notes into a timeline and a parts list. End with the sentence the manufacturer still needs. Do not submit the claim. ## Prompt You are working as Claude Cowork on one playbook. Task: Turn my notes into a timeline and a parts list. End with the sentence the manufacturer still needs. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not submit the claim. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Turn my notes into a timeline and a parts list. End with the sentence the manufacturer still needs. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not submit the claim. # Summarize a youth sports schedule URL: https://aiplaybookgems.com/usecases/gemini-summarize-a-youth-sports-schedule Agent: Gemini Category: Operations Roles: Parent Industries: Education Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://gemini.google/overview/ Gemini handles this as a draft: Practices and games in one table. Note conflicts with the family calendar I paste. Do not RSVP. ## Prompt You are working as Gemini on one playbook. Task: Practices and games in one table. Note conflicts with the family calendar I paste. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not RSVP. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Gemini where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Practices and games in one table. Note conflicts with the family calendar I paste. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not RSVP. # Summarize loyalty changes that hit my account URL: https://aiplaybookgems.com/usecases/openclaw-summarize-loyalty-changes-that-hit-my-account Agent: OpenClaw Category: Data Roles: Operator Industries: Hospitality Difficulty: easy Cost bucket: free Estimated cost: No paid plan is required to attempt the playbook. Model or hosting fees can still apply. Time saved: Ten minutes Status: unverified Source: https://openclaw.ai/ OpenClaw handles this as a draft: From the email, what changed and the date. Do not redeem points. ## Prompt You are working as OpenClaw on one playbook. Task: From the email, what changed and the date. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not redeem points. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open OpenClaw where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: From the email, what changed and the date. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not redeem points. # Turn a discharge note into a home schedule URL: https://aiplaybookgems.com/usecases/claude-cowork-turn-a-discharge-note-into-a-home-schedule Agent: Claude Cowork Category: Operations Roles: Operator Industries: Healthcare Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://claude.com/docs/cowork/overview Claude Cowork handles this as a draft: Meds and follow-up dates printed on the note. Do not change the plan. ## Prompt You are working as Claude Cowork on one playbook. Task: Meds and follow-up dates printed on the note. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not change the plan. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Claude Cowork where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Meds and follow-up dates printed on the note. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not change the plan. # Turn a whiteboard photo into a process list URL: https://aiplaybookgems.com/usecases/zapier-agents-turn-a-whiteboard-photo-into-a-process-list Agent: Zapier Agents Category: Operations Roles: Operator Industries: Software Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Twenty minutes Status: unverified Source: https://zapier.com/agents Zapier Agents handles this as a draft: Numbered steps. Mark unreadable words. Do not publish the process. ## Prompt You are working as Zapier Agents on one playbook. Task: Numbered steps. Mark unreadable words. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not publish the process. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Zapier Agents where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Numbered steps. Mark unreadable words. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not publish the process. # Turn interview tape notes into a scene list URL: https://aiplaybookgems.com/usecases/manus-turn-interview-tape-notes-into-a-scene-list Agent: Manus Category: Research Roles: Operator Industries: Agencies Difficulty: medium Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: An hour Status: unverified Source: https://manus.im/docs/introduction/welcome Manus handles this as a draft: Scene and the quote. Do not publish the piece. ## Prompt You are working as Manus on one playbook. Task: Scene and the quote. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not publish the piece. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Manus where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Scene and the quote. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not publish the piece. # Turn irregular freelance income into a monthly average URL: https://aiplaybookgems.com/usecases/hermes-agent-turn-irregular-freelance-income-into-a-monthly-average Agent: Hermes Agent Category: Data Roles: Founder Industries: Agencies Difficulty: easy Cost bucket: low Estimated cost: One individual paid plan is usually enough for a short run. Time saved: Half an hour Status: unverified Source: https://hermes-agent.nousresearch.com/docs/ Hermes Agent handles this as a draft: Use the deposits I label as income. Show the average and the worst month. Do not give tax advice beyond the arithmetic. ## Prompt You are working as Hermes Agent on one playbook. Task: Use the deposits I label as income. Show the average and the worst month. Rules: - Use only sources I give you. If a file, page, or number is missing, write unknown. - Do not invent URLs, prices, or names. - Stop rule: Do not give tax advice beyond the arithmetic. Return the work product first, then a short list of what you did not do because of the stop rule, and the single decision I still have to make. ## Steps 1. Open Hermes Agent where the source material already lives: Use the account, repo, or folder that has the real files. If you cannot see them, stop. 2. Paste the prompt and name the exact input: Use the deposits I label as income. Show the average and the worst month. 3. Check three facts against the source: A wrong date, amount, or name means the draft is not ready. 4. Keep the stop rule: Do not give tax advice beyond the arithmetic.