Prompts that teach you the mechanic

Summarize everything that happened while I was out since last Thursday: emails that need a reply from me specifically, decisions made in my Teams chats, and documents shared with me. Group by urgency, not by source. For each item, tell me the one action you'd recommend.

I'm meeting with /[person] in 30 minutes. Pull together: our last 3 email threads, any open commitments either of us made (search Teams and email), and the latest version of any file we've both edited. End with 3 questions they're likely to ask me.

Find where we decided [decision topic]. Search my emails, Teams messages, and meeting transcripts from the last 90 days. Quote the exact message or transcript line, who said it, and the date. If you find conflicting decisions, show me both.

Draft my weekly status update for my manager. Source it from: meetings I attended this week, emails I sent, and files I edited. Format: 3 wins, 2 risks, 1 ask. Write it in first person, confident but factual, max 150 words. Don't invent anything — if a section is thin, say so.

Draft a 2-page project proposal for [project]. Use /[requirements doc] for scope and /[budget spreadsheet] for the numbers. Match the tone and structure of /[a past proposal that won]. Flag any place where the source documents disagree with each other.

Rewrite this section at half the length without losing any commitments, numbers, or named people. Then list what you removed in bullet form so I can veto any cut.

Rewrite this document for a board audience: remove implementation detail, lead with financial impact, keep every number, convert any jargon to plain English, and keep it under one page. Preserve all headings as navigation anchors.

Add a column that flags any row where [metric] is more than 2 standard deviations from the mean of its [category] group. Explain the formula you used in plain English before applying it.

Convert this range to a formatted table, give every column a descriptive header, and tell me which columns have mixed data types or blanks that will cause problems for analysis.

Using advanced analysis, find the 3 strongest drivers of [outcome column] in this table. Show your Python, state your confidence in each finding, and tell me what additional column would most improve the analysis.

Create a presentation from /[Word doc]. Max 8 slides, one idea per slide, speaker notes containing the supporting detail you cut from the slide. Use our template's layouts only — no AI-generated images.

This slide has too much text. Split it into 2 slides with a clear narrative break, convert the bullets into a comparison layout if the content supports it, and rewrite each bullet to max 8 words.

Summarize this thread. Then list: 1) what's being asked of me specifically, 2) deadlines mentioned anywhere in the thread, 3) the position of each participant in one line each. Draft a reply that addresses only #1.

Draft a reply declining this request. Tone: warm but unambiguous. Offer one concrete alternative (suggest [alternative]). Don't apologize more than once, and don't leave a door open you don't intend to walk through.

What has been decided so far in this meeting, and what questions were raised but not answered? List who owns each open item based on what was said.

For the [meeting name] I missed: what was decided, what was assigned to me or my team, what was said about [topic I care about], and was there any disagreement? Quote the transcript for anything assigned to me.

Search this channel for everything posted about [project] in the last month. Build a timeline of status changes, and flag the messages where the stated deadline or scope changed.

You are helping me write instructions for a declarative Copilot agent. Its job: [job]. Its knowledge: [SharePoint site / files]. Draft instructions that include: persona, step-by-step response procedure, 3 concrete examples of good answers, what to do when the answer isn't in the knowledge source, and topics it must refuse. Keep under 8,000 characters.

Here are my agent's instructions: [paste]. Act as a hostile tester. Give me 10 user messages likely to make this agent fail, hallucinate, or leak content it shouldn't — then rewrite the instructions to survive all 10.

Rewrite the executive summary so every claim is supported by something in the body of the document. List any claim you couldn't support — those are either cuts or things I need to add evidence for.

Document this workbook for a colleague taking it over: what each sheet does, which cells contain manual inputs vs formulas, the data flow between sheets, and anything fragile that breaks if rows are added.

I'm new to [team/project]. Based on files, Teams channels, and meetings I have access to: who are the key people and what does each own? What decisions were made before I arrived that I should know? What's currently blocked and why?

Go through today's unread email. Tell me only the ones that: need a decision from me, contain a deadline, or come from my management chain. Everything else, give me a single combined one-paragraph digest.

Before creating slides: propose 3 different narrative structures for presenting [topic] to [audience], each as a 6-line story arc. I'll pick one, then you build the deck from it.

List every document in /[SharePoint site] that is older than 18 months, contradicts a newer document, or has 'draft' in the name or title. This is the cleanup list before I point an agent at this site.

Compare the action items from last week's [meeting] recap against what was actually posted or completed in this channel since. List each item as: done, in progress with evidence, or silent. Draft a gentle nudge for the silent ones.

Here's a decision we're about to make: [decision]. Search our files and threads for evidence that contradicts it — past attempts, failed pilots, dissenting opinions in Teams, data that points the other way. Steelman the case against.

Using advanced analysis on this table: model what happens to [output metric] if [input] drops 10%, 25%, and 40%. Present as a small table with the assumption stated above it, and tell me which assumption in my data the result is most sensitive to.

Catch me up on everything since yesterday 5pm, then organize my inbox: file newsletters and automated notifications away, flag anything needing a decision from me, and draft replies for the two most urgent threads — but show me each draft for approval before anything is sent or moved.

Plan our [team offsite / customer workshop] on [date] for [n] people: find a slot that works for the attendee list, draft the invite, create an agenda document and a logistics checklist in the output folder, and schedule a prep meeting with [co-organizer] two weeks before. Pause and ask me before sending any invites or booking anything.

Create a skill for my weekly status report. Interview me about: where the source material lives, the exact format my manager expects, my hard rules (never invent items, max 150 words, always show me the draft before sending), and what a thin week should look like. Then write the SKILL.md with a description that triggers whenever I ask for my 'Friday update'.

Research [competitor] as a threat to our [product line]. Sections: their offering and pricing, recent moves (last 12 months), where they win against us per our internal win/loss notes, where we win, and a one-paragraph recommended response. Sources: web for their public moves, our files and meeting transcripts for the win/loss view. Cite every claim. When you ask clarifying questions, I'll give you our current hypothesis to test against.

Before we commission outside research on [topic]: build a report of everything our org already knows about it. Search files, meeting transcripts, emails, and Teams across everything I can access. Sections: prior work and pilots, who the internal experts are, conclusions reached and by whom, open questions, and contradictions between sources. Quote and link every source.

Prepare a decision memo on [decision]. Structure: context, the case for, the strongest possible case against (steelman it — find real internal evidence: failed past attempts, dissenting voices in meetings or threads, data pointing the other way), risks of each path, and a recommendation with stated confidence. Web sources for market facts, work content for our history. Every claim cited.

Analyze this file for anomalies: outliers, impossible values, duplicates, sudden trend breaks, and any column where the distribution changed partway through. Show me the Python for every check, tell me which rows you excluded and why, and rank the anomalies by how much they'd change a topline summary of this data.

Using this dataset, identify the strongest drivers of [outcome] and compare cohorts by [dimension, e.g. signup quarter]. For each finding: show the Python, state your confidence as high/medium/low with the reason, flag any data-quality issue that could invalidate it, and tell me the one additional column that would most improve the analysis.