TL;DR: agents-md-facts is a small CLI that authors a minimal AGENTS.md from what your tree
already declares — real build/test commands, entry points, toolchain conventions. Nothing
invented. Nothing padded.
Why it exists
Hand-written agent files rot. LLM-written ones often bloat — and can reduce task success when the model pads them with plausible rules the agent then over-obeys.
This tool does neither. It reads the repo and writes only facts (or short curated defaults that resolve real ambiguity). Re-run when the code moves.
Four commands
npx agents-md-facts - (default) — author / refresh
AGENTS.md(non-destructive) --check— exit 1 if missing or stale (CI / pre-commit)--dry-run— print the plan; write nothing--stdout— emit the managed block to stdout
Only the block between <!-- agents:from-facts:start --> … <!-- agents:from-facts:end --> is managed. Everything you write outside those markers is preserved.
Keep it true in CI
Fail the build when the file drifts from the repo:
npx agents-md-facts --check Or pin the composite Action: uses: Wolfe-Jam/agents-md-facts@v0.1.0
Try it
No global install required:
npx agents-md-facts Source & stars: github.com/Wolfe-Jam/agents-md-facts · npm: agents-md-facts@0.1.0
The numbers
- v0.1.0 — first public release (2026-07-19)
- 4 commands — default ·
--check·--dry-run·--stdout - MIT · zero runtime dependencies · Node ≥ 18
- Examples — Node/TS · Python · Rust · Go
