ai-dev-kit
AGENTS.md1 starsChanged 26 days agoMITView on GitHub
Rules an agent must hold that the tree doesn't self-evidence; everything else is pointed at, not restated here. - Source vs install: edit skills/ and hooks/, never the .claude/ copies — the tracked .claude/ is the kit's own dogfood install. After any behavior change, re-run node install.mjs --adapter adapters/ai-dev-kit.json --hooks (add --global when a dual-home skill changed) or CI's root --check fails. - Behavior change ⇒ version bump: the kit version plus each touched skill's version in manifest.json, VERSION, the…
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 31,542 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 repository20 other files it gives its agents.
- checkpointSkill
- dep-checkSkill
- doc-auditSkill
- harness-auditSkill
- live-verifySkill
- project-adoptSkill
- project-auditSkill
- project-initSkill
- retroSkill
- tidySkill
- checkpointSkill
- dep-checkSkill
- doc-auditSkill
- harness-auditSkill
- live-verifySkill
- project-adoptSkill
- project-auditSkill
- project-initSkill
- retroSkill
- tidySkill
The file
We split this AGENTS.md into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 27
Gate before commit: run the adapter gate array (adapters/ai-dev-kit.json) and keep skill bodies generic — project facts go in an adapter or …
≈ “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
ai-dev-kit — agent onboarding
Rules an agent must hold that the tree doesn't self-evidence; everything else is pointed at, not restated here.
- Source vs install: edit
skills/andhooks/, never the.claude/copies — the tracked.claude/is the kit's own dogfood install. After any behavior change, re-runnode install.mjs --adapter adapters/ai-dev-kit.json --hooks(add--globalwhen a dual-home skill changed) or CI's root--checkfails. - Behavior change ⇒ version bump: the kit version plus each touched skill's
version in
manifest.json,VERSION, theCHANGELOG.mdtop entry, both deck stamps, and.claude-plugin/plugin.jsonmove together — CI gates the six sites. - Program discipline: forward-only banded backlog (
docs/BACKLOG.md); every row goes plan → explicit sign-off → build; a test that claims to catch a bug is shown failing first; CHANGELOG entries end with a Verification paragraph. - Hooks advise by default: an advisory handler exits 0 and injects context;
the agent decides. Blocking (exit 2) exists only in the enforcement handlers,
each inert without its adapter
enforcementkey. Exec-form wiring (command: "node"+ one anchoredargsentry) is a smoke-enforced invariant. - Zero dependencies: pure Node ≥ 22 — no package.json, no npm packages, no shell-specific scripts.
- Never rename a skill directory without a migration plan — stale-prune covers manifest-listed names only; a silent rename orphans every consumer's installed copy.
- Gate before commit: run the adapter
gatearray (adapters/ai-dev-kit.json) and keep skill bodies generic — project facts go in an adapter or project memory, never hardcoded in a skill. - A release commit is self-contained: version bump, self-install, and
fixture/doc updates land in one commit — the pre-tag gate checks only
the sha it is pointed at (why:
README.md→ Rules → Versioning). - One release commit per push, and let CI go green on it before the
next —
release.ymltags againstVERSIONas of the pushed sha, so a second bump in the same push silently skips the first release.
Pointers: README.md (status doc · install · release ritual) ·
CONTRIBUTING.md (local suite · ground rules) · docs/PLAYBOOK.md (the
why-layer) · manifest.json (machine index: versions · pipeline · hook
registry · decision log) · docs/BACKLOG.md (pending work).
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.

