| name | large-pr |
| description | Structure large cross-cutting changes into coherent phases with explicit dependency mapping and test checkpoints. Use when a feature touches multiple runtime subsystems or would otherwise produce an unreviewable single commit. |
Windsor Large PR
Apply when
- A feature affects multiple subsystems (runtime, evaluator, secrets, terraform, cmd, docs).
- A clean multi-PR split would require temporary churn or reversions.
Required planning output
Produce a change map before any implementation:
| Subsystem | Reason for change | Risk | Verification |
|---|
| ... | ... | ... | ... |
Implementation phases (single PR, multiple logical commits)
- Contract/test baseline — define behavior intent with stubs or failing tests.
- Core semantics — implement the primary logic change.
- Lifecycle/orchestration updates — wire new behavior into runtime flow.
- Integration wiring and security policy — persistence, secrets, permission checks.
- Boundary cleanup and style normalization — apply STYLE.md, remove speculative code.
- Documentation updates — update relevant docs and comments.
Control rules
- Avoid speculative refactors not required for feature correctness.
- If a refactor is required, tie it to a boundary or risk in the change map.
- Keep each phase independently testable.
- Report what changed and why after each phase.
Final review checks
- No boundary regressions introduced.
- No hidden global side effects.
- Public/private visibility is intentional.
- Files follow STYLE.md structure and comment rules.
task scan passes with no unresolved High findings.
task test:all passes.