Ask for work artifacts, not advice. Fill in the worksheet, copy a prompt, and replace the brackets with real information. Delete any field that adds nothing. The model only sees what you paste. It has not opened the web, your CRM, or your files.
Five-field worksheet
Use this before every prompt.
- Outcome: the artifact you need (email, memo, table, SOP, minutes).
- Audience: who will read it or who owns the decision.
- Context: facts, excerpts, and numbers you can stand behind.
- Constraints: length, tone, claims you must not make, and format.
- Quality check: how you will judge whether the output is usable.
If the first pass is wrong, do not ask the model to “make it better.” Quote the failing lines, say what is wrong, and ask for that revision only. Treat citations, figures, and proper names as unverified until you compare them with your documents or with the person accountable for the work. If you did not paste a source, the model has not checked it.
1. Writing and editing
1. Rewrite for a named reader
Use it when: the text is usable, but it is not written for the person who has to read it or decide.
Rewrite the source text for the named reader. Keep the meaning and the facts.
Purpose of the rewrite: [how the text will be used]
Reader: [role, what they already know, what you should not explain]
Source text:
[paste]
Facts that must not change: [names, figures, commitments]
Constraints: max [n] words; register [formal / plain / internal]; do not add examples, figures, or promises that are not in the source.
If a passage needs a missing fact, do not invent it: leave [MISSING: the fact needed] and continue.
Output:
1. Rewritten text
2. Substance changes (bullets; skip style notes)
3. Open questions
Check: that a new fact did not slip in while a sentence was being “clarified.”
2. Clarity edit that cuts words
Use it when: the draft already says what it needs to say, but it is long, repetitive, or hard to parse.
Edit the text so it is clear on first read and fits the limit. Do not change the thesis or the facts.
Purpose: version ready to [send / paste into a doc / read aloud]
Text:
[paste the draft]
Constraints: max [n] words; keep [figures, names, legal or scope caveats]; do not soften warnings.
If a sentence is ambiguous and the text does not resolve it, flag it and offer at most two readings. Do not pick one.
Output:
1. Edited text
2. Cuts that change emphasis (if any)
3. Sentences that are still ambiguous
Check: that a number, caveat, or scope limit did not get cut.
3. Send-ready email from notes
Use it when: you have loose points and need an email with a subject, a request, and a next step.
Turn the notes into a send-ready email. Do not add context that is not in the notes.
Purpose of the email: [inform / request / confirm / decline]
From: [name and role]
To: [name, role, relationship]
Notes:
[paste bullets, threads, or clippings]
Constraints: [register]; max [n] words; subject under [n] characters; do not offer [discounts, deadlines, formal apologies] unless they are in the notes.
If a fact required for the ask is missing (date, link, amount), do not invent it: leave [MISSING] and write the rest.
Output:
- Subject
- Body
- Facts still needed before sending
Check: that courtesy did not bury the actual request.
4. Structured document from a messy draft
Use it when: you have disordered prose and need a document with sections someone else can use.
Turn the draft into a structured document. Do not fill gaps with plausible content.
Purpose of the document: [what the reader should be able to do]
Reader: [role]
Draft:
[paste]
Desired structure: [e.g. context, decision needed, facts, risks, next steps]
Constraints: do not create empty sections; do not invent evidence headings; max [n] words.
If a section cannot be filled from the draft, leave [MISSING: what is needed] or drop it and say so.
Output:
1. Document with headings
2. Sections omitted, and why
3. Facts to request
Check: that whole sections did not appear from nowhere.
5. Change the tone, keep the facts
Use it when: the content is right and the register is not (too sharp, vague, internal, or salesy).
Rewrite tone only. Every claim must be traceable to the original.
Target tone: [plain / formal / neutral / internal]
Reader: [who will read it]
Text:
[paste]
Constraints: do not add emotion, urgency, praise, or apology; do not turn a hypothesis into a fact; do not drop figures or conditions.
If the original is vague, do not make it precise: keep the vagueness or mark [AMBIGUOUS].
Output:
1. Text in the requested tone
2. Sentences whose meaning may have shifted
Check: that the tone pass did not change what is being claimed.
6. Executive summary of a long document
Use it when: someone short on time needs what is decided, what is open, and what the text does not cover.
Summarize the text for a reader who will not read the original. Separate facts, inferences, and gaps.
Purpose: [prep a meeting / brief [role] / decide whether to read the source]
Decision owner (if any): [role]
Text:
[paste or excerpt]
Constraints: max [n] words; no outside knowledge; do not “round” figures.
If the text does not cover a point the purpose requires, call it a gap. Do not fill it.
Output:
1. What the text states (facts)
2. What can be inferred (labeled as inference)
3. What is missing for the purpose
4. The open question or decision
Check: that inferences are not mixed in with facts.
7. Consistency pass for names, terms, and style
Use it when: a document mixes product names, job titles, capitalization, or definitions.
Run a consistency pass. Do not “correct” a proper name that is used stably unless the glossary contradicts it.
Purpose: inconsistency list plus a harmonized text
Glossary or canonical form (if any): [term = correct form]
Text:
[paste]
Constraints: do not restyle beyond names, terms, dates, and number format; do not merge facts that only look similar.
If there is no canonical form and variants exist, pick the most frequent, say so, and list the others.
Output:
1. Table: term found | proposed form | count
2. Harmonized text
3. Cases you did not unify, and why
Check: that a valid proper name was not “fixed.”
2. Research and analysis
8. Source-cited summary of pasted material
Use it when: you need a summary that can be audited against the excerpts you actually have.
Summarize from the pasted sources only. Every finding must cite a source and a locator.
Purpose of the summary: [the question to answer]
Reader: [role]
Sources (label them A, B, C…):
[paste excerpts with title, author or URL if you have it, and a locator: page, paragraph, or date]
Constraints: do not cite sources that were not pasted; do not invent pages, dates, or quotation marks; if you paraphrase, say so.
If the sources do not answer the question, say so. Do not fill with general knowledge.
Output:
1. Short answer to the question
2. Findings with citations (Source X, locator)
3. Disagreements among sources
4. What the sources do not cover
Check: that no citation points to a document you did not paste.
9. Supplier comparison table
Use it when: you have proposals or notes from several vendors and need the same criteria applied to each.
Compare the suppliers using only the supplied data and the criteria I list.
Decision this prepares: [choose a supplier for…]
Decision owner: [role]
Criteria (in this order): [e.g. annual price, lead time, SLA, data exit, checkable references]
Suppliers:
[paste name + excerpts from each proposal]
Constraints: do not score a criterion with no data; do not use brand reputation or rumor; do not name a winner if material data is missing.
If a data point is missing, cell = [MISSING] plus a note on what to request.
Output:
1. Table: criterion × supplier (literal data or MISSING)
2. Material differences
3. Open questions per supplier
4. A recommendation only if every critical criterion has data; otherwise “no basis to recommend”
Check: that nobody was scored on a criterion you did not give.
10. Findings, gaps, and questions from a file
Use it when: you have a pile of notes, tickets, or PDFs and need what is known, what is not, and what to ask.
Extract findings and gaps from the file. Do not complete the story.
Purpose: [e.g. prep a review, a memo, or a call]
File:
[paste notes, emails, tickets]
Constraints: every finding needs an anchor (document or date); do not rank unless I give criteria; do not assume motives.
If two notes conflict, do not pick a side: record both.
Output:
1. Facts with anchors
2. Contradictions
3. Gaps (the missing fact and why it matters for the purpose)
4. Concrete questions for [role or person]
Check: that gaps were not filled with a plausible explanation.
11. Analytical read of a pasted table
Use it when: you have a table or numeric extract and need patterns, anomalies, and the limits of what can be claimed.
Analyze only the pasted table. Show the arithmetic. Do not extrapolate beyond the period or the sample.
Analysis question: [what we need to know]
Who will use this: [role]
Table:
[paste columns, units, and rows; say if rows were omitted]
Constraints: do not silently correct figures; do not invent totals if rows are missing; do not assign causes.
If a conclusion needs a fact that is not here (seasonality, population, metric definition), state that as a limit.
Output:
1. What can be claimed, with the calculation
2. Anomalies (outliers, holes, odd units)
3. What the table cannot support
4. Checks a human should run against the original source
Check: that figures were not rounded or recalculated without showing the original.
12. Synthesis of sources that disagree
Use it when: two or more documents conflict and you need the disagreement visible, not averaged away.
Synthesize the disagreement. Do not reconcile what the texts do not reconcile.
Question: [the disputed issue]
Sources:
[paste labeled excerpts]
Constraints: do not declare a winner; do not use outside information; do not smooth different figures.
If one source is opinion and another is a measurement, label them that way.
Output:
1. Points of agreement (with citations)
2. Points of disagreement (A says… / B says…, with citations)
3. Causes of disagreement that are visible in the text (definition, date, scope)
4. The data that would settle the dispute
Check: that a real contradiction was not averaged into a false middle.
13. Argument map from the evidence you have
Use it when: you will defend or challenge a claim and need arguments, counterarguments, and holes, not a speech.
Build an argument map from the pasted evidence only.
Claim: [the statement under review]
Decision owner: [role]
Evidence:
[paste excerpts]
Constraints: do not invent studies, analogies, or precedents; label each piece as fact, inference, or the author’s judgment.
If there is no evidence for an obvious counterargument, call it a hole. Do not fill it.
Output:
1. For (premise → cited evidence)
2. Against (premise → cited evidence)
3. Holes
4. The weakest part of the claim given this evidence
Check: that an author’s opinion was not treated as a proven fact.
3. Planning and decisions
14. One-page decision memo
Use it when: someone has to choose and needs a short memo with options, risks, and an explicit ask.
Write a decision memo. Separate facts, assumptions, and the recommendation.
Decision to take: [one sentence]
Decision owner: [role]
Deadline: [if any]
Context and facts:
[paste notes, figures, constraints]
Options already on the table (if any): [A, B, C]
Constraints: max [n] words; do not invent costs, timelines, or precedents; do not present the recommendation as inevitable.
If a material fact is missing, still deliver the memo and set the recommendation to “insufficient basis.”
Output:
- Context
- Decision requested
- Options (facts per option)
- Risks and reversibility
- Recommendation or “insufficient basis”
- Ask of the owner (approve / reject / request a fact)
Check: that the recommendation does not outrun the context.
15. Options memo with scored criteria
Use it when: several paths exist and you want them against the same criteria, with weights only if you supply them.
Compare the options against my criteria. Do not invent weights.
Decision: [what is being chosen]
Owner: [role]
Options:
[description or excerpt of each]
Criteria: [list]
Weights (if any): [criterion = weight; if I give none, do not score, only compare]
Constraints: empty cell = [MISSING]; do not use unlisted criteria; do not name a winner if a critical criterion is empty.
Output:
1. Table: criterion × option
2. Trade-offs
3. Least reversible option
4. Recommendation only if critical criteria are covered
Check: that weights or scores you did not assign did not appear.
16. Work plan with milestones and owners
Use it when: you have a goal and need an actionable plan, not a list of intentions.
Propose a work plan from the goal and the given constraints. Do not assign owners I did not name.
Goal: [observable result]
Deadline: [if any]
Who is commissioning the plan: [role]
People or roles available: [list]
Context:
[scope, dependencies, what is already done]
Constraints: do not invent tools, budgets, or “typical” durations; if you estimate, mark ESTIMATE and the assumption.
If a milestone has no owner, leave [NO OWNER] instead of assigning “the team.”
Output:
1. Milestones (result, date or sequence, owner)
2. Dependencies
3. Schedule risks
4. First action this week
Check: that dates and owners were not invented as if already agreed.
17. Prioritize a list of tasks or initiatives
Use it when: there is more work than capacity and you need an order based on explicit criteria.
Prioritize the list with my criteria. Do not use rhetorical urgency.
Capacity: [e.g. 3 items this week / 20 hours]
Ordering criteria, in this order: [impact, effort, risk, dependency…]
List:
[paste items with whatever detail you have]
Constraints: do not merge distinct items; do not invent effort if I did not give it; explain ties, do not break them on taste.
If an item cannot be scored, class it “no data” and do not weave it in as if it were scored.
Output:
1. Proposed order (one-line reason each)
2. Items beyond capacity
3. Items with no data
4. What would change the order if one assumption fails
Check: that the ranking is not based on what merely sounds important.
18. Go / no-go recommendation with an explicit bar
Use it when: you must continue, stop, or narrow, and you want the bar in writing before the verdict.
Issue a go / no-go / conditional-go verdict using the bar I define.
Decision: [ship, sign, spend, publish, hire…]
Owner: [role]
Go bar (required): [conditions that must be true]
Facts:
[paste evidence]
Constraints: do not move the bar; do not treat an assumption as a fact; do not add “momentum” or market analogies.
If the bar cannot be evaluated, the verdict is “not evaluable” and you list the missing fact.
Output:
1. Verdict
2. The bar, point by point: met / not met / not evaluable
3. Conditions if this is a conditional go
4. First action given the verdict
Check: that the analysis did not hide a recommendation by moving the bar.
19. Contingency plan for one named risk
Use it when: a risk has a name and you need triggers and responses, not a generic crisis plan.
Write a contingency plan for ONE risk. Do not cover risks I did not name.
Risk: [description]
Risk owner: [role]
Signals or triggers we already know: [list]
Real resources: [people, budget, systems]
Constraints: do not invent vendors, insurance, or recovery times; every action needs an owner from the list or [NO OWNER].
If there is no observable trigger, propose how it would be detected and mark that none exists today.
Output:
1. Trigger
2. Actions in the first [n] hours
3. What is communicated, to whom, using which fact
4. What not to do
5. Facts still needed before the plan is executable
Check: that the plan is not a generic crisis kit detached from the named risk.
4. Meetings and communication
20. Meeting agenda with a purpose and timeboxes
Use it when: you are calling a meeting and need a purpose, timings, and prep, not a topic list.
Write an agenda that can actually finish in the time given.
Meeting purpose (one result): [decision, alignment, review]
Meeting owner: [role]
Duration: [minutes]
Attendees: [roles]
Topics and pre-read material:
[paste notes]
Constraints: every block has minutes and an owner; the total cannot exceed the duration; do not add “AOB” that eats time; do not invite roles I did not list.
If a topic does not fit, move it to “not this meeting” with a follow-up owner.
Output:
1. Purpose in one sentence
2. Pre-read (if material exists)
3. Blocks: minutes | topic | owner | expected decision
4. What is out of scope
Check: that the agenda cannot be finished in the timebox.
21. Meeting minutes with decisions and owners
Use it when: the meeting has happened and you need what was agreed, not a transcript.
Write minutes from the notes. Separate decisions, actions, and topics that were only discussed.
Meeting: [name, date, owner]
Attendees: [list]
Notes or transcript:
[paste]
Constraints: do not turn a debate into a decision; do not assign actions to people who were absent unless the notes say so; do not invent dates.
If a “decision” has no owner or done-criterion, class it as open.
Output:
1. Decisions (who, what, by when)
2. Actions
3. Discussed, not decided
4. Next checkpoint
Check: that items only discussed were not recorded as decisions.
22. Status update for a named channel
Use it when: you need to report in Slack, email, or a weekly note, with the same facts for everyone.
Write a status update that matches the facts. The traffic light cannot be greener than the data.
Channel: [email / Slack / report]
Audience: [role]
Period: [dates]
Facts:
[progress, blockers, figures, dates]
Constraints: do not say “on track” if a blocker is open; do not invent ETAs; max [n] words.
If a milestone’s status is unknown, say unknown, not “in progress.”
Output:
- Traffic light per milestone (green / amber / red) with the justifying fact
- Done this period
- Blockers (owner, what is needed)
- Next visible milestone
Check: that the status does not sound better than the inputs.
23. Prep for a difficult conversation
Use it when: you have to deliver news, ask for a change, or set a boundary, and you want a script anchored in facts.
Prepare the conversation. Do not concede anything I did not authorize.
Purpose of the conversation: [what must be agreed]
Who is speaking, and with whom: [roles]
Facts that may be used:
[paste examples, dates, figures]
Line that must not be crossed: [e.g. do not offer more time / do not admit legal fault]
Constraints: do not invent the other person’s motives; do not add threats or flattery; do not propose compensation I did not list.
If a statement lacks a fact, remove it or mark it unusable.
Output:
1. Opening (facts)
2. Request or boundary, in one sentence
3. Three likely reactions and a factual reply to each
4. Close with a next step
5. Phrases to avoid
Check: that the script does not give away points you did not authorize.
24. Written briefing for one stakeholder
Use it when: someone short on time must walk into a meeting or a decision with the minimum that is sufficient.
Write a briefing. The decision or the risk comes first.
Stakeholder: [role, what matters to them as I described it]
Decision or question they will face: [text]
Facts:
[paste context]
Reading time: [e.g. 90 seconds]
Constraints: do not bury bad news; do not add recommendations unless I ask; do not use internal jargon without a gloss.
If you do not know what they will ask, list at most three likely questions labeled as hypotheses.
Output:
1. What they need to know now
2. The decision or help being asked of them
3. Risks and dates
4. Optional detail appendix
Check: that the decision needed is not buried under background.
25. Post-meeting follow-up with owners and dates
Use it when: you need to close a meeting in writing: what was agreed, who does what, and when.
Write the follow-up email or message. Only actions that appear in the notes.
Recipients: [list]
Meeting notes:
[paste]
Constraints: do not assign work to people who were not there unless the notes say so; do not bring forward the next date if none exists; register [specified].
If an action has no date, leave [NO DATE] and ask for one.
Output:
- Subject
- Decisions
- Actions (owner, deliverable, date)
- Open questions
- Facts still needed to close
Check: that work was not assigned to someone who was not in the room.
5. Sales and marketing
26. First-touch outreach using only supplied facts
Use it when: you are writing a first email or LinkedIn note and do not want invented personalization.
Write a first-touch message using only the facts supplied. No fictional research.
Purpose: [ask for a [n]-minute meeting / ask for a yes-no / share one fact]
From: [name, role, company]
To: [name, role, company]
Verified facts about them or their company:
[only what I paste: a page, a note, an event, a prior email]
Real offer or reason: [what you propose, no metaphors]
Constraints: do not invent their achievements, hires, pains, or figures; do not use urgency, discounts, or “I saw that”; do not claim customer results I did not document.
If there is not enough fact to personalize, use an honest generic reason and say so.
Output:
- Subject or first line
- Body
- Facts used (list)
- Facts I was tempted to invent and did not use
Check: that personalization was not guessed from the web or the industry.
27. Sales follow-up after a meeting or silence
Use it when: there has already been contact and the next note must rest on what was said, not on pressure.
Write a follow-up that matches the thread. One next step, not an ultimatum.
Thread context:
[paste emails, meeting notes, dates]
Purpose of this message: [answer X / propose a slot / send what was promised]
Constraints: do not invent “only a few spots left,” discounts, or expiry dates; do not recap benefits that were never mentioned; do not pretend they replied if they did not.
If there is no prior commitment, ask for a new one without attributing intent.
Output:
- Subject
- Body
- Prior commitments cited
- Proposed next step
Check: that fake urgency or a discount you did not offer did not creep in.
28. Content brief for one channel and one piece
Use it when: someone has to write or design a piece and needs a brief they can execute.
Write a content brief. Do not add keywords, audience claims, or promises I did not give.
Piece: [article, landing, thread, one-pager, email]
Channel: [where it runs]
Purpose of the piece: [what it must achieve, measurable if I gave a metric]
Audience: [the one I described]
Product or offer facts:
[paste]
Constraints: do not invent SEO, search volume, or testimonials; do not ask for a “disruptive” tone; include what not to say.
If the real CTA is missing, leave [MISSING] instead of inventing a sales close.
Output:
1. Core message (one sentence)
2. Usable facts
3. Piece structure
4. CTA
5. Forbidden claims
6. Definition of “ready”
Check: that audience or ranking claims were not added without evidence.
29. Proposal outline from discovery notes
Use it when: you left a discovery call and need the skeleton of a proposal, not the pitch.
Build a proposal outline from the discovery notes only.
Client or account: [name]
Discovery notes:
[paste]
Scope we can actually offer (if any): [list]
Constraints: do not invent price, timeline, team, or case studies; do not turn a vague comment into a requirement; mark assumptions.
If a critical requirement is missing, leave scope [INCOMPLETE] and list closing questions.
Output:
1. Problem in the client’s words (quoted)
2. Proposed scope vs. out of scope
3. Assumptions
4. Questions still needed before quoting
5. Proposal section structure
Check: that scope or pricing that was not in the notes did not appear.
30. Product or offer copy from verified facts
Use it when: you need product, one-pager, or page copy and can only claim what is documented.
Write the offer copy from product facts only. No testimonials and no unsourced figures.
Piece: [one-pager, page section, proposal paragraph]
Reader: [role]
Product facts:
[what it does, limits, price if any, requirements, what it does not do]
Constraints: avoid hype and transformation metaphors; do not invent comparisons; do not promise savings.
If a benefit has no supporting fact, do not write it.
Output:
1. Copy
2. Each sentence mapped to a fact (or removed)
3. Claims someone might want to add that are not supported
Check: that metrics, guarantees, or customer quotes did not appear unless they were in the facts.
31. Customer case write-up from supplied evidence
Use it when: you have permission and proof of real work and need a case, not a rounded story.
Write a customer case from the pasted evidence only. If a result is missing, the case goes without that result.
Permission to name the customer: [yes / no; if no, use [CUSTOMER]]
Evidence:
[problem, work done, dates, figures, verbatim quotes if any]
Case audience: [role]
Constraints: do not invent quotation marks; do not infer ROI; do not claim causality (“Y rose because of X”) unless the evidence says so.
If you only have an anecdote and no measurement, label it as an account, not a result.
Output:
1. Situation
2. Work done
3. Result (evidenced only)
4. Quotes (verbatim only)
5. Gaps that would block publishing
Check: that results or quotes were not fabricated.
32. Objection-response bank grounded in facts
Use it when: the team hears the same objections and needs answers that product and legal can stand behind.
Build objection responses. Each answer must rest on a fact I give.
Objections actually heard:
[paste]
Usable facts (product, price, process, limits):
[paste]
Constraints: do not deny a real limit; do not promise unlisted roadmap; do not contradict legal; do not say “everyone.”
If an objection has no supporting fact, give a clarifying question, not a rebuttal.
Output:
Table: objection | response | supporting fact | what not to say
Check: that the answers do not promise what product or legal cannot keep.
6. Operations and automation
33. As-is process map
Use it when: you need the process as it runs today (steps, roles, systems, exceptions), not the ideal version.
Map the process AS-IS. Do not redesign it.
Process: [name]
Process owner: [role]
Current description:
[paste notes, tickets, described screenshots]
Constraints: do not turn a “should” into a current step; do not drop waits or rework if they appear; do not add unmentioned tools.
If a step is unclear, draw it as [AMBIGUOUS] and write the question for the owner.
Output:
1. Trigger
2. Numbered steps: actor | action | system | output
3. Exceptions
4. Wait times mentioned
5. Questions to validate the map with the owner
Check: that a to-be process is not being presented as as-is.
34. Automation triage: what to automate, what not to, and in what order
Use it when: you have a list of repeated tasks and must decide what to automate, what to leave, and in what order.
Triage automation candidates. Rank with my criteria, not by technical novelty.
Criteria, in this order: [volume, clear rule, cost of error, build effort]
Candidate tasks:
[paste description, volume if any, systems, whether human judgment is required]
Constraints: do not recommend automating a step that needs unspecified judgment; do not invent ROI; do not assume APIs I have not confirmed.
If there is no volume and no rule, class it “not triable,” not “quick win.”
Output:
1. Automate now
2. Automate later
3. Do not automate (and why)
4. Missing data per candidate
Check: that a step still needing human judgment was not slated for automation.
35. SOP for a repeatable procedure
Use it when: work repeats and you need a procedure someone else can follow, including exceptions.
Write an executable SOP. Include the happy path and the exceptions I give.
Procedure name: [text]
Who runs it: [role]
When it triggers: [condition]
Known steps and exceptions:
[paste]
Real systems and permissions: [list]
Constraints: do not invent screens, fields, or approvers; every step needs an observable result; do not use “use common sense” as a step.
If a frequent-failure exception is missing, create the section and mark it [MISSING] instead of omitting it.
Output:
1. Purpose and scope
2. Roles
3. Numbered steps
4. Exceptions and escalation
5. Definition of done
6. Gaps that block publishing
Check: that the SOP does not skip the failure path.
36. Handoff checklist between roles
Use it when: work changes owner and data, files, or acceptance criteria get lost.
Write a handoff checklist. The last mile has an owner too.
From: [role A] → To: [role B]
Work being handed:
[paste]
Known acceptance criterion: [text]
Constraints: do not assume folders, tickets, or SLAs I did not mention; do not leave “coordinate” with no artifact.
If it is unclear who confirms receipt, leave [NO CLOSE OWNER].
Output:
1. Artifacts that travel (name, format, where they live)
2. Minimum data
3. Acceptance criterion
4. Who confirms receipt, and how
5. Typical failures of this handoff (only if they are in the context; otherwise leave the section empty)
Check: that final close has an owner.
37. Incident runbook for a known failure
Use it when: a failure has already happened or is foreseeable and you need detection, containment, and communication steps.
Write a runbook for this incident. Do not generalize to other failures.
Incident: [symptom]
Severity definition we use: [if any]
Systems: [list]
What was done or is known:
[paste]
Who can act: [roles]
Constraints: do not invent commands, dashboards, or on-call numbers; do not ask to restart anything I did not authorize; every step needs an expected result.
If there is no “resolved” criterion, mark [MISSING] and do not improvise one.
Output:
1. How it is detected
2. Immediate containment
3. Diagnosis
4. Communication (who, when, which fact)
5. Resolution criterion
6. What not to do
Check: that steps do not assume tools or access you never mentioned.
38. Definition of a recurring workflow
Use it when: a weekly or monthly ritual (close, reporting, review) currently lives in one person’s head.
Define the recurring workflow so it can run without the person who currently remembers it.
Workflow name: [e.g. weekly pipeline close]
Cadence: [weekly / monthly + day]
Owner: [role]
What is done today:
[paste]
Real capacity: [hours or people]
Constraints: do not propose a denser cadence than capacity allows; do not add new tools; do not turn a wish into a step.
If the workflow depends on a late-arriving input, document it as a dependency. Do not ignore it.
Output:
1. Trigger and cadence
2. Steps, owners, inputs, outputs
3. Definition of complete
4. Dependencies
5. What to cut if time runs out
Check: that the cadence ignores the capacity you stated.
7. Customers and team
39. Customer reply with facts and a next step
Use it when: there is a ticket, complaint, or question and you need a reply that does not over-promise.
Write the customer reply. One observable next step. No unauthorized compensation.
Channel: [email / chat / ticket]
Customer: [name or account]
Customer message:
[paste]
Usable internal facts:
[real status, real timelines, what can actually be done]
What must not be offered: [refunds, fault, dates]
Constraints: [register]; do not admit legal negligence; do not name-blame another person unless I ask; do not use reverse urgency.
If there is no real date, do not write “soon”: say what will happen and when the next update will come.
Output:
1. Send-ready reply
2. Promises made (so operations can see them)
3. What was deliberately left unsaid
Check: that an unauthorized apology-as-admission or a make-good did not slip in.
40. Role scorecard
Use it when: you are hiring, reshaping a seat, or aligning expectations: mission, results, and evidence.
Build the role scorecard from what I describe only. Do not copy a generic job ad.
Role: [internal title]
Mission of this seat in this company: [why it exists]
Results that are actually expected:
[paste]
Constraints: do not invent years of experience, tools, or fashionable soft skills; every result needs observable evidence; max [n] results.
If there is no definition of a good year, leave that cell [MISSING] instead of decorating it with a KPI.
Output:
1. Mission (3-4 lines)
2. Results (what / evidence / this is not)
3. Competencies only if they follow from the results
4. Interview questions anchored to each result
Check: that competencies were not lifted from a generic job ad.
41. 30/60/90 onboarding plan
Use it when: someone is joining and you need a plan with an owner, access, and a learning check, not a welcome list.
Build a 30/60/90 plan. Every block has an owner and a check.
Incoming role: [title]
Manager: [role]
Role context:
[paste scorecard, systems, team]
Real access and people: [list]
Constraints: do not invent trainings, buddies, or tools; do not load 20 tasks into week 1; every milestone needs “how we will see it is done.”
If an access item has no owner, [NO OWNER] and do not treat it as done on day 1.
Output:
- Day 1 (access and first deliverable)
- Days 30 / 60 / 90: goals, owner, evidence
- Onboarding risks
- First calibration conversation (date or Week n)
Check: that the 90-day plan has an owner and a proof of completion.
42. Synthesis of customer or team feedback
Use it when: you have surveys, 1:1s, or tickets and need themes with counts, not one vivid quote.
Synthesize the feedback. Separate frequency from intensity.
Source: [survey / interviews / tickets]
Material:
[paste]
Question to answer: [e.g. what is blocking renewal]
Constraints: do not treat an extreme quote as the majority; do not invent percentages that cannot be counted; do not mix customers and team if they come from different sources.
If n is small, say so and do not generalize.
Output:
1. Themes (count or n of mentions)
2. Verbatim quotes, max [n], labeled as illustration
3. Disagreements
4. What the material does not cover
Check: that an outlier quote is not presented as the majority view.
43. 1:1 agenda and talking script
Use it when: you have a 1:1 and do not want it to become a status dump.
Prepare the 1:1. Status gets a short block; the rest is manager and report work.
Roles: [manager and report]
Duration: [minutes]
Prep notes:
[paste]
Constraints: do not improvise feedback I did not write; do not turn the 1:1 into a backlog review; reserve minutes for [career / blockers / feedback] as I specify.
If there is no development topic, leave the block and note that prep is missing. Do not fill it with motivation.
Output:
1. Opening (2 min)
2. Status (facts)
3. Blockers
4. Feedback (only what I provided)
5. Development
6. Agreements and dates
Check: that the 1:1 did not collapse into a status dump.
44. Complaint or escalation reply
Use it when: the case has left the normal queue and you need a reply with facts, limits, and an internal owner.
Write the escalation reply. Separate internal and external.
Complaint:
[paste]
Internal facts:
[timeline, what was done, what can be done]
Internal owner: [role]
What legal or leadership authorizes saying: [text]
Constraints: do not admit unlawfulness; do not name-blame a vendor or a colleague; do not promise legal or refund timelines that are not authorized.
If the timeline has a hole, do not smooth it: mark [DATE GAP].
Output:
1. Customer-facing message
2. Internal note (timeline, risk, owner)
3. Promises made
4. Questions for legal, if any
Check: that a “helpful” reply does not contain a legal admission.
8. Quality, risk, and verification
45. Red-team review of a plan, contract, or draft
Use it when: a plan, contract, or public draft needs someone to try to break it before it goes out.
Act as red team. Rank ways this fails, not style nits.
Artifact:
[paste the plan, contract, or text]
Intended use: [who uses it, for what]
Owner: [role]
Constraints: do not rewrite the text except for a short counterexample; do not invent laws or incidents; every attack needs the assumption that makes it possible.
Ignore nits (commas, tone preference) unless they change an obligation.
Output:
1. Three most serious failure modes (impact × plausibility given this data)
2. Unsupported claims
3. Perverse incentives or exploitable ambiguity
4. Questions the owner should resolve before sending
Check: that real attack surface was not replaced by style nits.
46. Calculation and unit check
Use it when: a memo, quote, or model contains arithmetic and you need the sums, units, and rounding checked.
Check the calculations. Do not accept a total because it “looks right.”
Material:
[paste figures, described formulas, and units]
Question: [which total or ratio to validate]
Constraints: redo each calculation step by step; do not silently correct; do not use exchange rates, tax, or inflation I did not give.
If a figure has no unit or source, mark it [NOT CHECKABLE].
Output:
1. Recalculated workings
2. Matches and mismatches (with the delta)
3. Unit or rounding problems
4. Figures a human must still compare with the original source
Check: that arithmetic that is still wrong was not marked confirmed.
47. Privacy review of a flow or text
Use it when: a form, blast email, SOP, or profile moves personal data and you need to see what is collected and on what basis.
Review privacy from the pasted flow or text. Do not give conclusive legal advice.
Flow or text:
[paste]
Data we believe we handle: [list if any]
Lawful basis we believe we use: [the one I name; if I name none, do not assume one]
Constraints: do not cite statute articles unless I paste them; do not assert “we are GDPR/CCPA compliant”; flag unnecessary data and undeclared destinations.
If the basis is missing, the finding is “basis not stated,” not “illegal” or “legal.”
Output:
1. Personal data visible in the flow
2. Purpose the text declares (or [NOT DECLARED])
3. Recipients and retention mentioned
4. Practical risks (over-collection, forwarding, free-text fields)
5. Questions for legal or the data owner
Check: that a legal basis nobody named was not assumed.
48. Claim-by-claim source verification
Use it when: a text makes claims and you want a verdict per sentence: supported, partial, unsupported, or not checkable.
Verify each claim against the pasted sources. With no pasted source, nothing is “verified.”
Text to audit:
[paste, or number the sentences]
Sources:
[paste]
Constraints: do not use general knowledge to “confirm”; do not mark a sentence true if the source is silent; quotations must match.
If the source is ambiguous, verdict = partial or not checkable, not true.
Output:
Table: claim | verdict (supported / partial / unsupported / not checkable) | anchor | note
At the end: claims that should leave the text before it is published
Check: that a claim was not marked verified when the source is silent.
49. Check a deliverable against a policy
Use it when: there is a policy, brand book, or contract, and a deliverable that should follow it.
Check the deliverable against the pasted policy. “Not mentioned” is not “compliant.”
Policy or clause:
[paste]
Deliverable:
[paste]
Compliance owner: [role]
Constraints: do not interpret spirit beyond the text; do not cite outside rules; every finding needs an anchor on both sides.
If the policy is ambiguous, say so. Do not clear it.
Output:
1. Meets (with anchor)
2. Does not meet
3. Not evaluable (the policy does not cover that point)
4. Minimum changes to align it
Check: that policy silence was not read as a pass.
50. Pre-send quality checklist
Use it when: the text already feels “done” and you still need a pass for figures, names, promises, audience, and sources.
Build a checklist applied to THIS text, not a generic list. Score it against the original.
Text:
[paste]
Audience and channel: [who / where]
Claims that must have a source:
[list if any]
Constraints: every item gets pass / fail / not evaluable; do not invent brand standards; do not give figures a free pass.
If a strong claim has no source in the text or in what I pasted, it fails.
Output:
1. Applied checks (item | verdict | evidence)
2. Failures that block sending
3. Lesser failures
4. What a human must still look at (names, amounts, dates, links)
Check: that the checklist does not rubber-stamp unsourced claims.