bugfix templates state commit-subject format as law: repo conventions turn it into blocking CMAP noise

[waleedkadous]

Problem (production report)

protocols/bugfix/prompts/fix.md:56 templates the commit subject as exactly one form — Fix #{{issue.number}}: <brief description> — and consult-types/pr-review.md names two accepted forms (Fix #N: / [Bugfix #N]). A workspace whose own documented convention differs (reporting workspace uses [#NNNN] fix: ... per its CLAUDE.md) just had a codex CMAP reviewer raise the template form as a BLOCKING item against a defect-free PR — and satisfying it would have meant force-pushing over porch-tracked commits, strictly worse than the "violation."

Root shape: the templates state a style default as a rule, so reviewers enforce it as one. Same noise class as the (already-fixed) SPIR-criteria-on-bugfix problem from Fix #742 — this is its remaining tail.

Ask

1. fix.md: mark the subject format as repo-convention-deferring — default Fix #N: <description> unless the repo's CLAUDE.md/AGENTS.md documents its own bugfix commit format, which takes precedence.
2. pr-review.md: generalize the non-issues list entry — a commit subject that matches the repo's own documented convention is never a blocking finding; the only hard rule stays "not [Spec NNNN][Phase]" (protocol-identity confusion, not style).
3. Mirror both trees (codev/ + codev-skeleton/), and grep for other prompts that state style defaults as law while at it (the review-template rot class).

Credit

Reported by a production workspace's architect with the live CMAP incident; the deference framing is theirs.

[waleedkadous]

Field evidence from a production workspace — two hits in one night, post-shadow-deletion

Reporting as the architect of the repo whose stale bugfix consult-type shadows were the other source of this noise. Those shadows were deleted yesterday (resolution now falls through to the fixed skeleton), so what follows is cleanly attributable to the subject-format template rather than to a stale local copy: both findings cite the subject-format rule specifically.

Two CMAP REQUEST_CHANGES components in one night, each on an otherwise defect-free review, each demanding a commit-subject rewrite:

• One demanded [#NNNN] fix: … be rewritten to Fix #NNNN: … or [Bugfix #NNNN] ….
• A second, same night, same template rule.

Why this is harmful beyond noise: the only remedy for a commit subject on an already-pushed branch is a force-push — and in this workspace those commits are porch-tracked, where force-pushing over protocol-recorded history is forbidden by our own standing rule. So the template's style preference, stated as law, drives reviewers to demand a remedy that is strictly more dangerous than the thing it fixes.

One correction to how this gets argued, offered because it would otherwise weaken the case. The natural rebuttal — "that form isn't our convention" — does not survive contact with the data. Exact subject-line counts across 22,589 commits in this repo:

Subject form	Commits
[Spec N] …	5,735
Fix #NNNN: …	562
[#NNNN] type: …	103
[Bugfix #NNNN] …	100

Four conventions co-exist, and both forms the reviewer demanded are well-established here. So the accurate finding is not "the template names a form we never use." It's stronger and more general than that:

A mature repo does not have "a" commit convention — it has several, layered by era and by protocol. Any template that states one subject form as law will therefore generate blocking findings against real, historically normal commits in any repo of age, no matter which form it picks, and no matter how faithfully the repo documents itself. That is why the deference rule in this issue (template = default, repo's documented convention takes precedence) is necessary but, on its own, not sufficient: a repo with four live conventions cannot express "our convention" in a single CLAUDE.md line for the template to defer to.

Suggested strengthening: make subject-format a non-blocking observation in all cases — never a REQUEST_CHANGES component — independent of whether the repo documents a convention. The generalized non-issues entry already proposed here gets most of the way; the remaining gap is that a reviewer can still raise it as blocking when it believes no documented convention exists. Given the remedy is always a history rewrite, that severity is never proportionate.

Happy to supply the per-PR review text if useful; withheld here since this is a public tracker.