Turn a zero-shot idea into a perfectly-working, thoroughly-tested, spec-driven agent. One intake round (which also collects the API keys into .env), then the agent-builder builds one phase at a time — autonomous within a phase, with a human testing gate between phases. Also used to add a new capability to an existing agent.
Diagnose and fix a problem in an existing agent — a bug description, a runtime error/stack trace, failing tests, or spec/code drift — then verify the fix. Calls the worker agents directly; runs autonomously to a verified result.
Reconcile spec and code so they match. Audits the whole tree for drift, brings code in line with the spec (spec wins), and verifies. Calls worker agents directly; runs autonomously to a CLEAN audit.