cleat

cleat your instructions once, for any agent.

Keep a repo's instructions in one AGENTS.md that every agent reads, with CLAUDE.md reduced to a pointer at it. cleat is hooks only: no slash commands, no skills, nothing resident in session context.

📖 Read the docs →

Repo layout, the just targets, and the conventions this codebase holds itself to are in AGENTS.md, the same file the agents read. Requirements are in SPEC.md, their coverage in STATUS.md.

Exercising it without a session

The CLI is the whole implementation; the hooks are three registrations pointing at it. Both halves run standalone.

Report a directory's findings:

scripts/cleat check ../some-repo
scripts/cleat check ../some-repo --json

Feed it a hook payload the way Claude Code would:

printf '%s' '{
  "hook_event_name": "PreToolUse",
  "session_id": "local",
  "tool_name": "Write",
  "tool_input": {"file_path": "/tmp/demo/CLAUDE.md", "content": "# Demo\n\nRun `just test`.\n"}
}' | python3 scripts/cleat hook

Watching the hooks fire

The hooks produce no terminal output of their own. Launch with --debug and look for the [DEBUG] Hook PreToolUse / Hook PostToolUse lines:

claude --debug