Summary
The Win32 Button subclass dogfood used commit messages like boundary(implement): T008 subclass install for individual implementation tasks. That worked mechanically, but the wording blurs true lifecycle boundary commits with task-progress commits inside the implementation phase.
Evidence
The dogfood run committed each task in a focused way, which is good. The concern is only taxonomy: boundary(...) may be too broad for per-task implementation commits that do not themselves close a human-judgment boundary.
Expected
Decide and document the canonical commit prefix for implementation task commits. Candidate shape:
implement: T008 subclass installtask(implement): T008 subclass install- reserve
boundary(<stage>): ... for commits that close or sync lifecycle boundaries
Acceptance Criteria
- Update lifecycle discipline / coordinator guidance with the chosen taxonomy.
- Ensure examples in prompts and docs do not encourage misleading boundary prefixes for ordinary task commits.
- Keep focused per-task commits as the desired behavior.
Summary
The Win32 Button subclass dogfood used commit messages like
boundary(implement): T008 subclass installfor individual implementation tasks. That worked mechanically, but the wording blurs true lifecycle boundary commits with task-progress commits inside the implementation phase.Evidence
The dogfood run committed each task in a focused way, which is good. The concern is only taxonomy:
boundary(...)may be too broad for per-task implementation commits that do not themselves close a human-judgment boundary.Expected
Decide and document the canonical commit prefix for implementation task commits. Candidate shape:
implement: T008 subclass installtask(implement): T008 subclass installboundary(<stage>): ...for commits that close or sync lifecycle boundariesAcceptance Criteria