harness
AGENTS.md1 starsChanged 8 days agoMITView on GitHub
A universal, stack-agnostic project setup harness and skill library. A library project. It ships as a reusable package, CLI or archive. - All code, identifiers, comments, documentation, commit messages and release notes are written in English. - User-facing strings follow the project's own localization approach. Never hardcode a second language in code; keep locales in sync. - Use short, imperative commit messages in English with a Conventional Commits prefix: - feat: short message - a new feature. - fix: short…
- Reads credentials
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 repository9 other files it gives its agents.
- templates/AGENTS.mdAGENTS.md
- Add skillSkill
- Commit and pushSkill
- GitHub PagesSkill
- Logo designSkill
- Project setupSkill
- READMESkill
- Release cutSkill
- Release summarySkill
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
23 rules. About 2 are common, 15 are distinctive, and 6 are very specific to this project.
≈2
Common · estimate
Many files like this have a rule of this kind.
≈15
Distinctive · estimate
This author's own choices. Worth a look.
6
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
- Distinctive: line 20
All code, identifiers, comments, documentation, commit messages and release notes are written in English.
- Distinctive: line 27
Use short, imperative commit messages in English with a Conventional Commits prefix:
- Distinctive: line 38
Keep the subject under 72 characters, lowercase, and without a trailing period.
- Very specific: line 40
Never add AI authorship or co-author trailers: no Co-Authored-By, no "Generated by ...", no tool, agent or session attribution.
- Distinctive: line 47
Run the declared checks before committing.
- Distinctive: line 48
Do not amend, rebase, squash or force-push existing commits unless the user explicitly asks for it.
- Distinctive: line 50
Commit and push only when the user asks.
- Very specific: line 57
Use Semantic Versioning and tag releases as vX.Y.Z.
- Distinctive: line 67
Never move, delete or reuse an existing release tag.
- Distinctive: line 80
Run the declared checks before declaring any task done.
- Very specific: line 86
Follow setup/SETUP.md to configure a target repository.
- Very specific: line 93
After changing templates, guidelines or scripts, run node tests/run.mjs and node tests/pilot.mjs, and keep every test green.
- Distinctive: line 95
Never let a template change break rendering for any of the five project types.
- Distinctive: line 96
Do not add stack-specific, framework-specific or architecture guidance here.
- Distinctive: line 98
Keep the renderer deterministic: no timestamps, no environment-dependent output, no network calls.
- Very specific: line 107
Before a task that matches a skill, read its SKILL.md and follow it:
- Common: line 117
Never include credentials, absolute paths or project-specific details in a skill.
- Very specific: line 121
Never commit secrets, API keys, tokens, credentials, .env files, local environment configuration, build output or local planning notes.
- Common: line 123
Never log secrets, prompts or private data.
- Distinctive: line 124
Never force-push and never rewrite published history unless the user explicitly requests it on a branch they own.
- Distinctive: line 126
Never invent data, sources or dependencies.
- Distinctive: line 129
Remove dead code, commented-out code and unused imports before committing.
- Distinctive: line 130
Do not add comments that restate the code.
≈ “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
Harness
A universal, stack-agnostic project setup harness and skill library.
Project type
A library project. It ships as a reusable package, CLI or archive.
Commands
| Purpose | Command |
| Check | node tests/run.mjs |
| Check | node tests/pilot.mjs |
| Check | node web/build.mjs --check |
| Release | gh workflow run release.yml -f bump=minor -f summary="..." |
Language rules
- All code, identifiers, comments, documentation, commit messages and release notes are written in English.
- User-facing strings follow the project's own localization approach. Never hardcode a second language in code; keep locales in sync.
Commits and pushes
Use short, imperative commit messages in English with a Conventional Commits prefix:
feat: short message- a new feature.fix: short message- a bug fix.refactor: short message- a change that neither adds a feature nor fixes a bug.chore: short message- maintenance that does not change behavior.docs:,test:,perf:,build:,ci:,style:- the remaining allowed types.
- Keep the subject under 72 characters, lowercase, and without a trailing period. Use the body only for context, wrapped at 72 columns.
- Never add AI authorship or co-author trailers: no
Co-Authored-By, no "Generated by ...", no tool, agent or session attribution. The author is the human contributor only. - Stage explicit paths. Never use
git add -A,git add .orgit commit -a. Never stage local planning files, secrets, environment files or build output. - Review the staged diff before committing.
- Run the declared checks before committing. A failing check blocks the commit.
- Do not amend, rebase, squash or force-push existing commits unless the user explicitly asks for it.
- Commit and push only when the user asks. When they do, commit the intended changes and push to the current branch.
Versioning and releases
- Releases are cut through the
ReleaseGitHub Action (.github/workflows/release.yml,workflow_dispatch). Never create release tags by hand. - Use Semantic Versioning and tag releases as
vX.Y.Z. Pre-releases use a suffix such asvX.Y.Z-alpha.1. - The version comes from the latest
vX.Y.Zgit tag; no file is bumped. - Choose a bump type (
patch,minor,major,prerelease,premajor,preminor,prepatch) or provide an explicit version. - Release notes are generated from Conventional Commits since the previous tag
by
scripts/release-notes.mjsusingdocs/RELEASE_TEMPLATE.md. - Pre-release versions contain a hyphen and are published as GitHub pre-releases.
- Describe user-visible changes. Do not list commits that are not shipped.
- Never move, delete or reuse an existing release tag. Fix problems with a new patch release.
- The first release of a new project is
v0.1.0.
CI and build
- Recorded CI depth:
ci. CIruns the project checks on ubuntu-latest, windows-latest.- Checks, in order:
node tests/run.mjsnode tests/pilot.mjsnode web/build.mjs --check- Run the declared checks before declaring any task done.
Operating this repository
- This repository is the harness: it configures other repositories and hosts the universal skill library. It does not contain application code.
- Follow
setup/SETUP.mdto configure a target repository. It is the only supported entry point. setup/decisions.mdis the locked contract. Change it only with the user, and update the templates and tests in the same change.setup/questions.mdis the question bank. Keep it in sync with the decisions.templates/holds deterministic assets;guidelines/holds composableAGENTS.mdsections;skills/is the skill source of truth.- After changing templates, guidelines or scripts, run
node tests/run.mjsandnode tests/pilot.mjs, and keep every test green. - Never let a template change break rendering for any of the five project types.
- Do not add stack-specific, framework-specific or architecture guidance here. Those belong in a future optional skill.
- Keep the renderer deterministic: no timestamps, no environment-dependent output, no network calls.
Skills
This repository is the source of the skill library. Skills live in
skills/<id>/SKILL.md; add or change them there and update
skills/README.md.
Before a task that matches a skill, read its SKILL.md and follow it:
commit-and-push- review, stage explicit files, write a Conventional Commit, commit and push.release-summary- write a short user-facing summary of the latest release.release-cut- trigger the release workflow, watch it and verify the result.github-pages- publish a static site or generated documentation to GitHub Pages.readme- create, standardize or translate a README.logo- design a logo system with five versions before creating SVGs.add-skill- propose a new reusable skill and apply it in the harness repository.
Never include credentials, absolute paths or project-specific details in a skill.
Secrets and safety
- Never commit secrets, API keys, tokens, credentials,
.envfiles, local environment configuration, build output or local planning notes. - Never log secrets, prompts or private data.
- Never force-push and never rewrite published history unless the user explicitly requests it on a branch they own.
- Never invent data, sources or dependencies. Verify that a dependency exists before using it, and cite the source of any external data.
- Handle errors explicitly. Never fail silently.
- Remove dead code, commented-out code and unused imports before committing.
- Do not add comments that restate the code. Use comments only for non-obvious decisions.
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.

