Getting started

crag is a CLI. You edit one governance file, AGENTS.md. crag keeps your CI gates, git hooks, and per-agent config files in sync with it. The package is @whitehatd/crag (v1.0.0-alpha.1). It has no runtime dependencies and needs Node 18 or newer. Installing it adds two binaries: crag and crag-mcp.

Install

npm install -g @whitehatd/crag
# or run it once, no install:
npx @whitehatd/crag

First run

Run crag with no arguments inside a repo. It runs analyze then compile in one pass.

cd your-project
crag

What lands on disk:

Step	What it does	Files written
`analyze`	Reads `package.json`, CI workflows, Makefiles, and 25+ other signals; infers your test/lint/build gates.	Updates the managed gates fence inside `AGENTS.md`. Scaffolds a minimal `AGENTS.md` if none exists.
`compile`	Detects which AI tools and CI systems are actually present, then derives the config each one needs from `AGENTS.md`.	`AGENTS.md` plus only the deltas (for example `CLAUDE.md`, Cursor rules, a CI gates workflow). Never all 23 files.

Check for drift

crag audit compares the committed governance against the repo as it exists now: stale compiled files, gates that reference a tool you no longer have, CI commands missing from AGENTS.md.

crag audit            # human-readable report
crag audit --fix      # recompile stale targets
crag audit --ci       # exit non-zero on drift (for pipelines)

Zero runtime dependencies. The published package.json lists no dependencies. crag runs on Node built-ins alone. Nothing is fetched at install beyond the package itself.