# 10. Recipes and pitfalls

- **Bug:** `poteto-mode` -> bug-fix -> optional `tdd` -> `interrogate` -> real-surface proof.
- **Architecture:** `architect` -> optional `arena`/`prototype` -> implement -> `blast-radius` -> verify.
- **Large migration:** multi-phase-plan -> `show-me-your-work` -> verifiable units -> session-pickup as needed.
- **Performance:** perf-issue or hillclimb -> measured baseline -> one hypothesis at a time -> measured acceptance.

Common failures: treating sequential role-play as independent review, optimizing without a baseline, preserving legacy APIs by reflex, loading every skill into context, claiming a proxy proves the real surface, and asking the user questions that a reversible experiment could answer.