agentic-playbook
AGENTS.md0 starsChanged 15 days agoMITView on GitHub
Don't assume. Don't hide confusion. Surface tradeoffs. Before implementing: - State your assumptions explicitly. If uncertain, ask. - If multiple interpretations exist, present them - don't pick silently. - If a simpler approach exists, say so. Push back when warranted. - If something is unclear, stop. Name what's confusing. Ask. - Do not factor in development cost. Minimum code that solves the problem. Nothing speculative. Ask yourself: "Would a senior engineer say this is overcomplicated?" If yes, simplify. Touch only…
- Installs packages
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository25 other files it gives its agents.
- templates/CLAUDE.mdCLAUDE.md
- a11y-auditSkill
- architecture-reviewSkill
- bootstrap-projectSkill
- capture-screenshotsSkill
- customizeSkill
- deliver-featureSkill
- design-reviewSkill
- docs-upkeepSkill
- fix-bugSkill
- handoffSkill
- implementSkill
- leadSkill
- log-decisionSkill
- refine-requirementsSkill
- release-readinessSkill
- reproduce-in-worktreeSkill
- security-auditSkill
- ship-prSkill
- start-taskSkill
- sync-branchSkill
- threat-modelSkill
- verify-featureSkill
- vet-dependencySkill
- write-adrSkill
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
21 rules. About 0 are common, 19 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈19
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 11
If multiple interpretations exist, present them - don't pick silently.
- Distinctive: line 12
If a simpler approach exists, say so.
- Distinctive: line 13
If something is unclear, stop.
- Distinctive: line 14
Do not factor in development cost.
- Distinctive: line 20
No features beyond what was asked.
- Distinctive: line 21
No abstractions for single-use code.
- Distinctive: line 23
No error handling for impossible scenarios.
- Distinctive: line 24
If you write 200 lines and it could be 50, rewrite it.
- Distinctive: line 26
Ask yourself: "Would a senior engineer say this is overcomplicated?" If yes, simplify.
- Distinctive: line 32
When editing existing code:
- Distinctive: line 33
Don't "improve" adjacent code, comments, or formatting.
- Distinctive: line 34
Don't refactor things that aren't broken.
- Distinctive: line 36
If you notice unrelated dead code, mention it - don't delete it.
- Distinctive: line 38
When your changes create orphans:
- Very specific: line 39
Remove imports/variables/functions that YOUR changes made unused.
- Distinctive: line 40
Don't remove pre-existing dead code unless asked.
- Distinctive: line 67
Ask product questions the ticket doesn't settle: 2–4 options, recommended first.
- Very specific: line 68
No AI attribution in commits, PRs, comments or docs (no Co-Authored-By, no "Generated with").
- Distinctive: line 73
Read the docs for the installed version of a fast-moving API.
- Distinctive: line 74
Stop the processes you started.
- Distinctive: line 116
Before calling a task done: <list the commands above that must pass>.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Working in this repository
<!-- The playbook's default AGENTS.md. Sections 1–7 are the rules for every project: keep 1–4 as they are, and adapt 5–7 if the project needs to. Section 8 is this project's own rules. The installer appends the chosen stack and tracker sections after it. Projects own their copy: edit it freely. -->1. Think Before Coding
Don't assume. Don't hide confusion. Surface tradeoffs.
Before implementing:
- State your assumptions explicitly. If uncertain, ask.
- If multiple interpretations exist, present them - don't pick silently.
- If a simpler approach exists, say so. Push back when warranted.
- If something is unclear, stop. Name what's confusing. Ask.
- Do not factor in development cost.
2. Simplicity First
Minimum code that solves the problem. Nothing speculative.
- No features beyond what was asked.
- No abstractions for single-use code.
- No "flexibility" or "configurability" that wasn't requested.
- No error handling for impossible scenarios.
- If you write 200 lines and it could be 50, rewrite it.
Ask yourself: "Would a senior engineer say this is overcomplicated?" If yes, simplify.
3. Surgical Changes
Touch only what you must. Clean up only your own mess.
When editing existing code:
- Don't "improve" adjacent code, comments, or formatting.
- Don't refactor things that aren't broken.
- Match existing style, even if you'd do it differently.
- If you notice unrelated dead code, mention it - don't delete it.
When your changes create orphans:
- Remove imports/variables/functions that YOUR changes made unused.
- Don't remove pre-existing dead code unless asked.
The test: Every changed line should trace directly to the user's request.
4. Goal-Driven Execution
Define success criteria. Loop until verified.
Transform tasks into verifiable goals:
- "Add validation" → "Write tests for invalid inputs, then make them pass"
- "Fix the bug" → "Write a test that reproduces it, then make it pass"
- "Refactor X" → "Ensure tests pass before and after"
For multi-step tasks, state a brief plan:
1. [Step] → verify: [check]
2. [Step] → verify: [check]
3. [Step] → verify: [check]Strong success criteria let you loop independently. Weak criteria ("make it work") require constant clarification.
5. Working agreement
<!-- Defaults: edit them for the project. Details: ${CLAUDE_PLUGIN_ROOT}/principles/00-working-agreement.md. -->- The owner decides product and merges every PR. Never merge or push to the default branch; branch first. Keep CI green.
- Ask product questions the ticket doesn't settle: 2–4 options, recommended first. Choose sensible defaults for the rest and say so.
- No AI attribution in commits, PRs, comments or docs (no
Co-Authored-By, no "Generated with"). - Secrets: never print, paste, log or commit them, and never read the keychain; the owner sets them (give the exact command). Only publishable keys reach a client.
- Vet each new dependency first (
vet-dependency); install with the ecosystem's version-aware installer. - Reproduce bugs and review findings in a git worktree and try the fix there before you involve a person; agents that change files work in their own worktree (
reproduce-in-worktree). - Confirm before anything hard to undo or outward-facing. Report results faithfully, including failures and what you didn't verify.
- Read the docs for the installed version of a fast-moving API; don't trust memory.
- Stop the processes you started; leave alone what you didn't start. No tight polling. Ask yourself: am I cutting corners, or drifting from what the customer needs?
6. Non-negotiables
- Data: migrations only, never edit a merged one; each new table gets a tenant id, deny-by-default access, grants and access tests; soft delete and purge; regenerate types.
- Code: catch an error only where you handle it (never swallow, re-throw unchanged or log twice); comments say why, never what (principle 01 §5, §7).
- Writes are idempotent; retries live in one layer; classify errors by code, never by message.
- Time: UTC instants plus an IANA zone; calendar dates as dates; no hand-rolled zone maths.
- UI: design tokens and the component kit; loading, error, offline and empty states; never show "empty" before data arrives.
- Copy: in the message catalogue (ICU), from the reader's side; never infer pronouns from a name.
- Accessibility: WCAG 2.2 AA (largest text size, 24 px minimum targets, labels, a cue besides colour, reduced motion).
- Privacy: collect the minimum; no private content in logs, analytics, crash reports or notifications.
7. Definition of done
The full check suite passes; you saw the change work in the real product, with evidence; the docs are updated in the same PR; the PR follows the template and the ticket has the verification.
${CLAUDE_PLUGIN_ROOT} is the playbook folder: the plugin's folder (the session-start hook prints it), or .playbook/ in a repo that keeps a copy for Cursor or Copilot. Principles: ${CLAUDE_PLUGIN_ROOT}/principles/; templates: ${CLAUDE_PLUGIN_ROOT}/templates/. A project's copy at .playbook/local/<same path> wins over the playbook file, and this file wins over both.
8. This project
<!-- Fill these in during bootstrap; delete the hints. Keep each rule one bullet, with the path to the code that implements it. -->- What it is: <one sentence: who it's for and what it does>. PRD:
<link>. - Tracker: <Linear / Jira / GitHub Issues / …>, project
<name or link>. Access, ids, status mapping and where the docs live are in the "Tracker" section below (installed with--tracker; for another tracker, copy one from.playbook/trackers/). Living docs: Decision Log<link>, Architecture<link>, Runbook<link>, PRD<link>. - Delivery: <PR (the default)> or <local merge: no PR; the owner squash-merges each branch locally>.
ship-prreads this line. - Stack (see
docs/adr/0001-stack.md): <framework + exact major version>,<backend>,<database>,<hosting>. Docs to check before using an API:<versioned docs URLs>. - Tenant unit: <workspace / account / space>. Every content row has
<tenant>_id.
Commands
| Task | Command |
| Install | <npm ci / pnpm i / …> |
| Run locally | <…> |
| Lint | <…> |
| Typecheck | <…> |
| Unit tests | <…> |
| DB / integration tests | <…> |
| Generate types | <…> |
| Add a dependency | <npm i / pnpm add / nx add / …> <pkg> |
Before calling a task done: <list the commands above that must pass>.
Layout
<src/app/>: <routes / entry points><src/components/>: the UI kit (preview at<…>)<src/data/>: the one client per service, queries, outbox<db/migrations/>,<db/tests/>docs/adr/: ADRs
Environments
| Env | Where | Deployed by |
| local | <…> | developer |
| dev | <…> | pipeline on merge |
| prod | <…> | manual pipeline run |
Domain rules
<!-- One numbered subsection per feature area as it's built, e.g. "### 9. Auth", "### 10. Billing". Each bullet: the rule, then the code path. -->Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

