agentic-foundry
AGENTS.md1 starsChanged 3 months agoMITView on GitHub
Claude Code reads this file as the project's instructions when there is no CLAUDE.md (since 2.1.278, 2026-09-18); other agent tools read it by convention. It holds only the invariant rules of this repo, each verifiable against the tree. Orchestration and session-specific prompts live elsewhere (skills/, the adopter's own workspace). This repo is a Claude Code plugin (foundry): skills in skills/*/SKILL.md, personas in agents/, hooks in hooks/ (wired by hooks/hooks.json), Python tooling in scripts/, tests in tests/, templates in context/, stack…
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 31,774 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 repository71 other files it gives its agents.
- llms.txtllms.txt
- amendSkill
- auditSkill
- authorize-releaseSkill
- authorizeSkill
- certify-localSkill
- certify-stagingSkill
- coherence-checkSkill
- command-deckSkill
- contextSkill
- cut-releaseSkill
- dashboards-as-codeSkill
- data-tier-cutoverSkill
- decommission-gateSkill
- deploy-statusSkill
- dispatchSkill
- doctorSkill
- drift-sweepSkill
- env-hygieneSkill
- extract-specSkill
- fleetSkill
- grounding-conformanceSkill
- id-applySkill
- id-architectSkill
- id-baselineSkill
- id-discoverSkill
- id-documentSkill
- id-driftSkill
- id-impactSkill
- id-implementSkill
- id-importSkill
- id-planSkill
- id-promoteSkill
- id-reviewSkill
- id-rollbackSkill
- id-simulateSkill
- id-syncSkill
- id-testSkill
- id-validateSkill
- id-verifySkill
- indexSkill
- infra-sandboxed-applySkill
- initSkill
- intakeSkill
- learn-captureSkill
- learn-distillSkill
- merge-when-greenSkill
- mode-autonomousSkill
- mode-interactiveSkill
- modeSkill
- post-upgradeSkill
- releaseSkill
- relockSkill
- report-citation-graphSkill
- reposSkill
- research-captureSkill
- research-firstSkill
- revertSkill
- sd-debugSkill
- sd-discoverSkill
- The first 60 shown.
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
Skill frontmatter is load-bearing: every skills/*/SKILL.md must keep YAML-parseable frontmatter with name and description.
≈ “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
AGENTS.md — the standing rules for coding agents working in this repo
Claude Code reads this file as the project's instructions when there is no CLAUDE.md
(since 2.1.278, 2026-09-18); other agent tools read it by convention. It holds only the
invariant rules of this repo, each verifiable against the tree. Orchestration and
session-specific prompts live elsewhere (skills/, the adopter's own workspace).
This repo is a Claude Code plugin (foundry): skills in skills/*/SKILL.md, personas
in agents/, hooks in hooks/ (wired by hooks/hooks.json), Python tooling in scripts/,
tests in tests/, templates in context/, stack profiles in packs/.
Ground rules
- Run the tests:
python3 -m pytest tests/ -q(full suite, ~10 min measured 2026-09-21; the certify-fixture module boots a localhttp.server) with the pinned dev requirements (requirements-dev.txt; CI uses Python 3.12), andpython3 scripts/foundry-doctor.py(must printDOCTOR-GREEN— run from this repo as the session root; run from a hosting workspace it reports the control-plane RED by design). Both must be green locally before the first push of a branch. - The hook's executable content is digest-pinned in
tests/fixtures/release/cut-release-reconcile-sweep.yaml; a deliberate change tohooks/foundry-git-discipline.shadvances the pin with a numbered ledger entry there. - Never weaken an assertion to make a test pass. Fix the code or flag the test.
- No claim beyond shipped enforcement: docs and skill prose must never describe guarantees the code doesn't provide. When you change behavior, change the prose in the same PR.
- Skill frontmatter is load-bearing: every
skills/*/SKILL.mdmust keep YAML-parseable frontmatter withnameanddescription; the doctor checks all of them. - Versions are pinned: GitHub Actions by 40-char SHA; dependency changes go through the standing-versions discipline — don't float anything.
- Git discipline: branch per change, PR-then-merge, no force-push to
main. The repo's own PreToolUse hook (hooks/foundry-git-discipline.sh) refuses force-pushes to a protected branch,gh pr merge --admin, and any merge ahead of green checks — by design, not a bug to work around. Atoms PR intorelease/<version>;mainreceives one PR per release;hotfix/<id>→mainis the one exception. Seedocs/how-to/branching-and-cleanup.md. - Text that names a guarded command goes through Write/Edit, never a Bash heredoc. The
hook scans the whole command string, heredoc bodies included, so
cat > f <<EOFwith a body mentioning a force-push or--adminis refused (docs, tests, commit bodies, PR bodies). An inline-m "…"is fine. Seedocs/merge-floor.md. - The retired-framework token fails CI.
ci.ymlgreps the whole tree, case-insensitive, for the three-letter name of the retired environment tool (the one that starts withcand ends withx); any occurrence in any file — code, comment, doc, fixture — is red. - PR trailers: code-change PRs carry a
Spec:trailer naming the authorizing spec (thespec-link-basegate in.github/workflows/btb-gates-base.ymlchecks it; docs-only diffs are exempt). Security-relevant changes (this hook, CI gate workflows, authorization, provenance) get an independent reviewer — self-review has missed real bugs here.
Where to learn more
README → docs/QUICKSTART.md → docs/architecture.md → docs/glossary.md → CONTRIBUTING.md. The changelog is the authoritative history.
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.

