Conversation
ssheld
left a comment
There was a problem hiding this comment.
🤖 Review by Claude Opus 4.6 · via Claude Code
Formal review state unavailable: cannot approve own pull request
No findings above the confidence threshold.
One line, exactly where we agreed it should go (Core Principles, after the planning guardrail), identically mirrored. This matches the #56 consensus plan — PR 1 of 2.
The wording is well-calibrated. "Do not widen scope into unrelated cleanup unless it was explicitly requested or is necessary for correctness or maintainability of the current change" gives two clear exception paths (user asked for it, or it's needed for the change to be correct/maintainable) without leaving room for "this nearby code could use some cleanup" drift.
Placement as the last bullet in Core Principles is correct — it's a scope discipline rule that applies across all work, not specific to any one workflow phase.
Strengths
- Minimal, focused change that delivers the highest-value import from #56.
- The two exception paths ("explicitly requested" and "necessary for correctness or maintainability") prevent the guardrail from being too rigid while still closing the opportunistic-cleanup loophole.
Merge Recommendation
Approve
Top Risks
- None.
Suggested Additional Tests
- None beyond the mirror check already run.
Summary
Files Changed
scaffold/agent-vault/shared-rules.md: add the new Core Principles guardrailscaffold/agent-vault/AGENTS.md: mirror the shared-rules changeValidation
bash scripts/check-policy-mirrors.shgit diff --checkRisks
Docs Consistency
README.mdordocs/design.mdchanges needed.Notes
Refs #56