pm-agentic-cookbook
AGENTS.md2 starsChanged 32 days agoApache-2.0View on GitHub
If your goal is to use this toolkit in another project, read AGENT_GUIDE.md. If that guide is absent, report that the usage guide is unavailable and do not infer installation support. The instructions below govern agents modifying this repository. Resolve the Definition, Provider and Consumer roles before changing a skill. Keep domain nouns out of the Definition; put domain-specific rules in an explicitly selected Provider. Consumers supply complete requests and host conventions. Keep execution-critical references inside each skill. Optional cookbook…
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 31,700 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 repository13 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- llms.txtllms.txt
- acceptance-criteriaSkill
- check-prdSkill
- consistency-auditSkill
- create-prdSkill
- domain-modelingSkill
- journey-auditSkill
- source-intakeSkill
- spec-captureSkill
- table-template-designerSkill
- test-authoringSkill
- ui-generatorSkill
The file
We split this AGENTS.md into the rules it gives an agent
9 rules. About 1 is common, 5 are distinctive, and 3 are very specific to this project.
≈1
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 3
If your goal is to use this toolkit in another project, read AGENT_GUIDE.md.
- Distinctive: line 4
If that guide is absent, report that the usage guide is unavailable and do not infer installation support.
- Distinctive: line 19
Keep domain nouns out of the Definition.
- Very specific: line 21
Keep execution-critical references inside each skill.
- Distinctive: line 24
Run evals from the source directory, not from installation packages.
- Very specific: line 28
Before committing, run npm run doc-sync and npm run test:gates and inspect the exact diff.
- Distinctive: line 48
Only the maintainer merges.
- Common: line 51
Keep changes surgical. Do not commit private data, credentials, external reference materials, or identifiable customer content.
- Distinctive: line 55
When authority, permissions, license ownership or scope is unclear, stop and ask.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 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
Contributor instructions
If your goal is to use this toolkit in another project, read AGENT_GUIDE.md.
If that guide is absent, report that the usage guide is unavailable and do not infer
installation support. The instructions below govern agents modifying this repository.
Repository map
process/: workflow stages, verification loops and executor responsibilities.skills/: independently usable instructions, references, templates and build scripts.specs/: a reference rule library selected through the consuming project's spec map.templates/: workflow-level document skeletons.examples/: synthetic or explicitly public-safe scenarios.tools/: gates, regression tests and advisory hooks.
Skill changes
Resolve the Definition, Provider and Consumer roles before changing a skill.
Keep domain nouns out of the Definition; put domain-specific rules in an explicitly
selected Provider. Consumers supply complete requests and host conventions.
Keep execution-critical references inside each skill. Optional cookbook pointers
must include a standalone-install fallback. Rebuild the skill's dist/ after
changing its body or references, and record the version in its CHANGELOG.
Run evals from the source directory, not from installation packages.
Validation and decisions
Before committing, run npm run doc-sync and npm run test:gates and inspect the
exact diff. Report each command's numeric exit code and every failure or skipped
check; if the tool output omits the exit code, capture it instead of inferring it.
Gates do not certify business acceptance. Hooks are advisory.
Propose reusable rules through spec-capture; the maintainer decides whether to
adopt them. Do not self-certify approval. Use an independent reviewer for changes
to audit skills: authors must not audit their own artifacts.
Language and translations
Reply in the user's language. Chinese documents are authoritative where an English
sidecar exists. Change Chinese first, synchronize the English text and update
source_version using git hash-object --no-filters. Never change only an English
rule. Agent-facing files are English-only and do not have sidecars.
Contribution and licensing
External contributors use a fork and a focused branch, then submit a PR to main.
Agents explicitly authorized by the maintainer may use task branches in this
repository, such as codex/<task>, claude/<task> or antigravity/<task>.
Only the maintainer merges. Use English Conventional Commit messages and English
branch names and PR text. Preserve human author identity. Obtain explicit user
authorization before commit, push or PR actions performed on their behalf.
Keep changes surgical. Do not commit private data, credentials, external reference
materials, or identifiable customer content. Preserve applicable copyright,
license and NOTICE information; mark modifications to redistributed files.
This repository uses Apache-2.0; separately licensed material retains its terms.
When authority, permissions, license ownership or scope is unclear, stop and ask.
Host adapters
CLAUDE.md and GEMINI.md import this file and contain only host integration
information. Keep each adapter at or below 40 lines and omit shared-rule headings.
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.

