Fourteen carefully engineered prompts and four templates that turn Claude into a reliable co-worker for the writing and thinking you do every week — meeting notes, weekly updates, decision memos, email, doc summaries, specs, SOPs, customer feedback, and research synthesis.
Not for coding. For the other 80% of knowledge work: the recurring documents that eat your afternoons. Built for PMs, founders, consultants, ops leads, chiefs of staff — anyone who runs on Claude but isn't shipping code with it.
Most people use Claude like a slightly-smarter search box: they type a vague request, get a generic wall of text, and quietly do it themselves anyway. The gap isn't the model — it's the prompt. A knowledge-work task has structure (a meeting has decisions, owners, and open questions), and a prompt that encodes that structure gets you output you can actually ship.
This pack is that structure, written down. Each prompt is engineered the way Anthropic recommends prompting Claude — a clear role, XML-tagged inputs, explicit output format, and guardrails against the specific way that task goes wrong (inventing owners, inflating status, hallucinating a competitor's pricing). You paste your mess in; you get a clean, honest draft out.
| # | Prompt | Turns… | …into |
|---|---|---|---|
| 1 | prompts/meeting-notes-to-actions.md |
raw notes / transcript | decisions, owned action items, open questions |
| 2 | prompts/weekly-update.md |
a Friday brain-dump | a crisp status update your manager reads |
| 3 | prompts/decision-memo.md |
"we need to decide X" | a one-page memo with a real recommendation |
| 4 | prompts/inbox-triage-email-draft.md |
a full inbox / one message | a triage table + a reply in your voice |
| 5 | prompts/summarize-long-doc.md |
a 40-page report or contract | an anchored, verifiable summary + watch-outs |
| 6 | prompts/competitive-brief.md |
scattered competitor notes | a strategy-ready brief (no invented facts) |
| 7 | prompts/project-kickoff-brief.md |
"we should do X" | a one-pager with goal + explicit non-goals |
| 8 | prompts/interview-synthesis.md |
raw user-interview notes | themes with quotes and honest frequencies |
| 9 | prompts/prd-drafter.md |
a rough "we should build X" | a scoped spec with explicit non-goals |
| 10 | prompts/status-rollup.md |
many teams' separate updates | one exec summary that leads with what's off-track |
| 11 | prompts/feedback-to-themes.md |
a dump of reviews / tickets / NPS | prioritized themes with honest counts + a next step |
| 12 | prompts/sop-writer.md |
a process that lives in your head | a repeatable SOP someone else can follow |
| 13 | prompts/meeting-prep-brief.md |
a meeting you haven't prepped | a one-screen brief with objections + answers |
| 14 | prompts/rewrite-for-audience.md |
a blunt or rambling draft | a retargeted version, same meaning, right tone |
Want a one-line "use this when…" for every prompt, template, and example in one
place? See INDEX.md.
Templates (templates/) — fill-in-the-blanks docs that pair with the prompts:
one-pager, weekly update, decision log, meeting agenda.
Examples (examples/) — three prompts shown on realistic messy input with the
output to expect, so you can see the quality before you buy your own time back.
Guides — START-HERE.md (5-minute setup) and GUIDE.md (how to get 2× more
out of every prompt: Projects, voice samples, verification habits).
Every prompt is built around one principle — an AI draft is only useful if you can trust it enough to send. So each one is wired to:
- Never invent accountability — missing owners/dates become
OWNER?/DATE?, not a plausible guess. - Separate fact from inference — what a document says vs what the model infers is always labeled.
- Refuse to inflate — status updates and memos stay factual; "refactored auth," not "dramatically improved security."
- Stay verifiable — summaries anchor every claim to a page/section so you can check before you rely.
That discipline is the product. The prompts are how it's delivered.
Any Claude surface — the Claude app, Projects, the API, or Claude in your editor. No account beyond Claude, no install, no subscription to this pack. Copy a prompt, paste your content, go. Provider-neutral in spirit, but tuned for Claude's strengths (long context, XML-tag steering, faithfulness to source).
Open START-HERE.md, do the 5-minute setup, and run prompt #1 on your last
meeting's notes. If it doesn't save you an hour in the first week, you weren't
going to use it anyway.
This pack is free and MIT-licensed and stays that way. Nothing in the prompts is crippled, time-limited, or holding back a "pro" version.
The same lab sells five paid packs. They are all for people working with AI coding tools, which is a different job from the document work this pack covers — so this is worth your money only if that is also you:
| AI Coding Workflow Pack — $24 | 26 files: 6 subagents (code-reviewer, debugger, test-writer, refactorer, doc-writer, pr-describer), 8 slash commands, and 6 per-stack rules files for TypeScript+React, Python, Go, Node APIs, Next.js and monorepos |
| Autonomous Agent Starter Kit — $29 | Templates, guardrails and field notes for running an AI coding agent unattended |
| AI Coding Security Pack — $29 | Subagents, commands and stack-specific rules that give an AI coding tool a dedicated security reviewer |
| Agent Constitution Pack — $19 | Seven complete, annotated CONSTITUTION.md governance files for autonomous-agent businesses |
| 18-Point Pre-Deploy Security Checklist — $1 | A compact editable checklist: secrets, config, auth, data handling, dependencies, infrastructure |
All at https://fablerlabs.com — card or crypto, instant download, no account.
There is also a second free one, the AI-Coding Field Guide and a
CLAUDE.starter.md, at https://fablerlabs.com/field-guide.
Made by Fabler Labs. These prompts and templates were
authored by an autonomous AI agent — they're engineered against each task's
failure modes, but every prompt still produces a draft: you're the editor, so
read it before you send it. MIT-licensed — see LICENSE.md. Use every file in
your own work, commercial included, no attribution required.