AGENTS.md
Ralph Loop that orchestrates Ralph Loops across a GitHub account/org.
STOP CONDITION
All target repos have functioning Ralph infrastructure installed and verified.
WHAT THIS REPO DOES
This is a fleet orchestrator. It:
- Scans GitHub repos (personal or org)
- Installs/updates Ralph infrastructure on each
- Reports fleet status
- Can trigger coordinated actions across repos
ABSOLUTE RULES
- PRD only: Work from
prd.json - One story per iteration
- Never commit secrets: GitHub tokens live in
.envonly - Verify before commit:
bun test && bun tsc - Dry-run first: Destructive operations require
--dry-runpass first
LOCAL DEV
# Install
bun install
# Set GitHub token
cp .env.example .env
# Add: GITHUB_TOKEN=ghp_xxx (needs repo scope)
# Run commands
bun run scan # List repos
bun run status # Fleet status
bun run install <repo> # Install Ralph on one repo
bun run sync # Sync config across fleet
GITHUB API
- Auth:
GITHUB_TOKENenv var - Rate limit: 5000/hr authenticated
- Use octokit:
import { Octokit } from '@octokit/rest'
RALPH INFRASTRUCTURE (what gets installed)
Target repos get:
AGENTS.md(from templates/)scripts/ralph/prd.json(initialized)scripts/ralph/progress.txt(empty).github/workflows/ralph.yml(optional CI)
KNOWN FOOTGUNS
- Don't overwrite existing AGENTS.md without
--force - Check for existing PRD stories before reinitializing
- Rate limiting: batch API calls, sleep between repos