Sync-to-Codebase

Derive a project's entire agentic setup from its code — memory files, a subagent fleet, hooks/ permissions, skills, MCP — persist it to disk, and keep it in sync as the code evolves. The full feature set is intact; the trick is that bundled scripts do the hard mechanical parts so even a small local model executes it reliably, one bite-sized task at a time.

The one idea

The codebase is the source of truth for what the config should be; .sync-codebase/ (disk) is the source of truth for what it currently is. You don't hold it all in your head and you don't hand-author fragile output. You break the work into tasks, do ONE at a time, RUN a script to do the mechanical work, COPY a template instead of inventing, write a short report, then VERIFY — and only then move on.

Golden rules (these prevent every known failure)