software-factory
AGENTS.md6 starsChanged 2 days agoMITView on GitHub
Software Factory is a TypeScript monorepo (npm workspaces under packages/) that ships GitHub issues autonomously through a boss-worker-checker pipeline: PLAN → BUILD → CHECK → SHIP. core is the UI-less engine, cli is the factory command, and config holds the typed defaults and constitutions. Step-by-step procedures live in .claude/skills/. Read the matching SKILL.md before you do the task.
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 33,315 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 repository5 other files it gives its agents.
- docs/agents.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- merge-policySkill
- run-evalsSkill
- verifySkill
The file
We split this AGENTS.md into the rules it gives an agent
7 rules. About 0 are common, 1 is distinctive, and 6 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
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
- Very specific: line 7
Run bash scripts/verify.sh --no-e2e and get it green before every commit.
- Very specific: line 8
Never lower a coverage threshold in vitest.config.ts.
- Very specific: line 9
Never merge past a required check that is genuinely FAILURE (gh pr merge --admin or equivalent).
- Very specific: line 10
Write or update the colocated *.test.ts next to every source change.
- Very specific: line 17
Do not write comments. anti-slop/no-comments allows only justified lint overrides (-- <reason>), rule notes in oxlint.config.ts and …
- Distinctive: line 18
Use domain terms (lane, park, claim, trust tier, ...) as defined in CONTEXT.md.
- Very specific: line 22
Do not "fix" the test script in packages/core/package.json.
≈ “Common” is an estimate: at least 1 in 5 of the 42,283 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
Software Factory is a TypeScript monorepo (npm workspaces under packages/) that ships GitHub issues autonomously through a boss-worker-checker pipeline: PLAN → BUILD → CHECK → SHIP. core is the UI-less engine, cli is the factory command, and config holds the typed defaults and constitutions.
Rules
- Run
bash scripts/verify.sh --no-e2eand get it green before every commit. Run it from the repo root. - Never lower a coverage threshold in
vitest.config.ts. When the ratchet says coverage rose, raise the threshold in the same PR. - Never merge past a required check that is genuinely
FAILURE(gh pr merge --adminor equivalent).mainmust always be green. - Write or update the colocated
*.test.tsnext to every source change. - Package boundaries (config ← core ← cli, zero-dependency config, published-name imports, ADR-0004 subpaths) are enforced by the
anti-slop/package-boundarieslint rule innpm run lint; its messages say how to fix a finding. - Model routing lives in
packages/config/src/defaults.ts. Do not hard-code model lists incore. core's root export is the narrow public API; see ADR-0004 and theanti-slop/package-boundariesrule for the/internaland/testingsubpaths.- TypeScript is strict and ESM only. Use
.jsextensions on relative imports. - Record significant design decisions as ADRs in
docs/adr/(see its README). - A user- or ops-visible change adds a
## Unreleasedentry toCHANGELOG.mdand updates the docs users read (README,docs/status.md,docs/agents.md,docs/models.md,docs/ops/) in the same PR. Never hand-editpackage.jsonversions. See docs/ops/docs-and-changelog.md. - Do not write comments.
anti-slop/no-commentsallows only justified lint overrides (-- <reason>), rule notes inoxlint.config.tsand toolchain directives, andcomments-baseline.jsonmay only shrink (ADR-0177). - Use domain terms (lane, park, claim, trust tier, ...) as defined in CONTEXT.md.
Known agent traps
- Do not "fix" the
testscript inpackages/core/package.json. It is intentionally"test": "vitest run". Tests only resolve from the repo root, where coverage is configured. Runnpm run testthere instead.
Skills
Step-by-step procedures live in .claude/skills/. Read the matching SKILL.md before you do the task.
verify— what the verify gate runs, how to read and fix each failure, coverage ratchet, integration tests.merge-policy— landing PRs, the two legitimate bypass cases, what to do whenmainis red.run-evals— stub, single-case, and full LLM-judge eval runs and baseline comparison.
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.

