debug

ALWAYS invoke when the user reports a bug, failing test, or wants to verify a hypothesis. Triggers: "debug X", "fix this bug", "why is X failing", "stack trace", "investigate X", "is X actually doing Y", "prove/disprove". Two modes via `certainty:` — high (default, full TDD fix) or low (hypothesis investigation, no fix). When NOT to use: new features (use build), quality polish on known-good code (use polish).