Playbooks turn recurring Python work into reviewable steps with an explicit definition of done.
| Task | Use it when | Proof of completion |
|---|---|---|
| Fix a bug with a regression test | Behavior is wrong or has regressed | Reproduction fails before the fix and passes after it |
| Add or change an API | A public HTTP or library contract changes | Contract, errors, tests, and compatibility are verified |
| Integrate an external API | Python calls a third-party service or model | Timeouts, failures, credentials, and test doubles are covered |
| Upgrade dependencies | Runtime or packages need updating | Lockfile, tests, advisories, and runtime smoke check pass |
| Ship a release | A package or service is ready to publish | Artifact, changelog, deployment, and rollback are verified |
| Write deterministic evals | Changing LLM prompts, models, or tool-calling agents | Golden datasets and deterministic assertions pass without regressions |
| Set up structured logging | Moving from scripts to production services | Structured JSON, trace ID propagation, and redaction are tested |
| Manage database migrations | Modifying SQL schemas or ORM models | Upgrade, downgrade, backfill safety, and zero-downtime compatibility pass |