AI agents don't fix a messy system — they multiply whatever structure already exists. In a multi-repo brownfield, the make-or-break work isn't in any one service: it's at the seams, and the seams live in the ★ glue repo that sets up and orchestrates the rest. Click around the map. Check off what your system already has. The meter tells you how much you can safely delegate.
Your checkmarks are saved in this browser only — nothing leaves the page.
— click a repository to open its prep list —
Make the implicit explicit. Every repo explains itself in a file agents read
(AGENTS.md); the glue repo explains the system. Every verification an
agent needs — build, test, lint, diagnose — is one command that works from a fresh clone.
Contracts between repos are committed artifacts a CI can diff, not tribal knowledge.
Guardrails are written down, and the machine enforces what matters. None of this is
AI-specific — it's onboarding hygiene with a deadline. The agent is just the new hire
who joins a thousand times a day, reads everything, and believes exactly what you wrote.
Distilled from converting a real multi-repo side project to agent-first development. No frameworks, no trackers — one HTML file. · more fun pages