agents-md

# agents-md

> agents-md generates and safely maintains an AGENTS.md file for any code repository.
> AGENTS.md is the open, cross-tool standard (backed by the Linux Foundation's Agentic AI
> Foundation) that AI coding agents read to understand a project: its commands, conventions,
> gotchas, and do-not-touch boundaries. agents-md ships as a zero-dependency `npx` CLI and as a
> Claude Code skill. It never fabricates (unknowns become stubs), and it never overwrites
> content a human wrote: re-runs only refresh regions wrapped in HTML-comment markers, and a
> handwritten AGENTS.md with no markers is left untouched (a AGENTS.generated.md is written
> instead).

## What it is
- A generator + maintainer for AGENTS.md, the "README for AI coding agents".
- Works across Codex, Cursor, GitHub Copilot, Gemini CLI, Aider, Windsurf, Zed, and more. CLAUDE.md is symlinked to AGENTS.md.

## Use
- CLI (any repo, no dependencies, no API key): `npx @eugeniughelbur/agents-md`
- Claude Code skill: `npx skills add eugeniughelbur/agents-md`, then run `/agents-md`

## Key links
- Repository: https://github.com/eugeniughelbur/agents-md
- README: https://github.com/eugeniughelbur/agents-md/blob/master/README.md
- The AGENTS.md template it produces: https://github.com/eugeniughelbur/agents-md/blob/master/templates/AGENTS.template.md
- AGENTS.md standard: https://agents.md

## Why it exists
Generic AI-context generators either fabricate details or overwrite hand-written docs on re-run.
agents-md is safe-by-default: deterministic detection, explicit stubs for unknowns, and
machine-enforced markers so re-running is always a refresh, never a reset. Its most valued output
is the three-tier Boundaries section (Always / Ask first / Never), which prevents destructive
agent mistakes.