pr-management-code-review: AI-attribution footer claims "confirmed by a Magpie maintainer" without verifying the runner is one

[choo121600]

What's broken

The pr-management-code-review skill appends an AI-attribution footer to every review it posts, asserting the submission was "confirmed by a Magpie maintainer" and that "a Magpie maintainer — a real person — will take the next look at the PR." The skill never verifies that the person running it actually holds a maintainer role, so this claim is emitted verbatim even when the runner is an ordinary contributor with no write/maintainer role on the target repo. The result is a public review body that makes an inaccurate factual claim about who vouched for it.

Which layer

• skills/pr-management-code-review/posting.md → § AI-attribution footer (all three variants — APPROVE / REQUEST_CHANGES / COMMENT — hard-code the "maintainer" wording).
• skills/pr-management-code-review/SKILL.md → Golden rule 5, which mandates the footer verbatim ("do not paraphrase, do not omit") and describes it as reassuring the contributor that "an maintainer — a real person — has confirmed the submission."
• Related gate: skills/pr-management-code-review/prerequisites.md → § 1 (the collaborator/permission pre-flight).

How to reproduce

1. As a GitHub user who is not a maintainer/committer on the target repo (a contributor only — no admin/maintain/write), run /magpie-pr-management-code-review pr:<N> against a public PR.
2. Accept a disposition and confirm posting. A COMMENT review succeeds from any account — GitHub lets any user submit a PR review on a public repo regardless of write access.
3. The posted review body ends with:

   This review was drafted by an AI-assisted tool and confirmed by a Magpie maintainer. … a Magpie maintainer — a real person — will take the next look at the PR.

Observed live on apache/magpie#902 (public), posted from a non-committer contributor account. The PR's reviewDecision stayed REVIEW_REQUIRED afterward — confirming the account is not a maintainer whose review satisfies the branch-protection gate, even though the footer told the contributor a "maintainer" had confirmed.

Expected vs actual

Expected — the attribution is accurate about who confirmed the review. Either:

• (a) the skill verifies the runner holds a maintainer/write role before it emits "maintainer" wording, or
• (b) the footer wording is role-neutral — e.g. "confirmed by the reviewer running Magpie" / "confirmed by a project maintainer or contributor" — so it never asserts a role the runner may not hold.

Actual — the footer asserts "maintainer" unconditionally. SKILL.md Golden rule 5 forbids paraphrasing or omitting it. And prerequisites.md § 1's permission probe (admin/maintain/write) both (1) conflates "write collaborator" with project "maintainer", and (2) per its own text only "warns and offers dry-run" rather than blocking — so a non-maintainer can post a review that publicly, and inaccurately, claims maintainer confirmation.

Surface area (optional)

• All three disposition templates inherit the same footer, so every posted review is affected.
• Adopters who install the pr-management-* family inherit this footer wording in their own repos, so the inaccuracy propagates to adopter projects.
• Likely fix touchpoints: derive the footer text from a verified role check in prerequisites.md, and relax Golden rule 5's verbatim constraint enough to allow the role-accurate variant while keeping the "AI-drafted, human-confirmed, may contain mistakes" substance intact.

Environment (optional)

• Harness: Claude Code (Opus 4.8), skill pr-management-code-review.
• Framework version: self-adopted framework checkout (.apache-magpie.lock → method: local, source: skills/); no snapshot drift.
• Observed target: apache/magpie (public), PR feat(codex): first-class Codex CLI skill runtime #902.