Repository navigation
fix(review): quality-gate pr mode assumes code-review always posts; it now outputs to the terminal by default #13010
Description
Problem
plugins/review/skills/quality-gate/context/pr.md treats the marketplace code-review plugin (/code-review:code-review) as a surface that always posts to the PR. It gates the plugin as a mutation (its Mutation gate) and never runs it as a read-only pass. The pointer record there carries the recheck trigger "a mode that returns findings to the session".
That trigger has fired. The plugin's official README (https://raw.githubusercontent.com/anthropics/claude-plugins-official/main/plugins/code-review/README.md, fetched 2026-11-04) says the command "Outputs review (to terminal by default, or as PR comment with --comment flag)" and documents --comment as opt-in.
Fix
- Re-read the plugin's README and
commands/code-review.md, and record what the default run does now. - Decide whether
quality-gate prmay run/code-review:code-reviewwithout--commentas a read-only pass, and whether the Mutation gate then covers only the--commentform. - Update
pr.md's pointer record (as-of date and trigger) and the mode's routing to match.
Acceptance
pr.mdno longer depends on the plugin having no terminal mode.- The Mutation gate's scope matches what the plugin's default run actually does.
Refs: #12697 (found while verifying the pointer swaps in that PR).
Activity
github-actions commentedon Nov 4, 2026
Proposed work class: work-class: scoped
Basis: automated intake triage; the lanes App applied this class.
Size: Medium
Complexity: Low
Uncertainty: Known
Flags: none
Blocked by: none
Understood: quality-gate's pr mode and the fanout skill assume
/code-review:code-reviewalways posts to the PR. The plugin now prints to the terminal by default and posts only with--comment.#13010 (comment)
The decision settles the four open questions, so nothing is asked again. No duplicate found.
Automated triage, applied by the lanes App.
melodic-automation-lanes commentedon Nov 4, 2026
Not admitted to the issue lane yet: no trusted-human acceptance criteria: a listed human must put the confirmed
- [ ]criteria in the body or post them as their own comment, or, on an issue a listed human filed, intake's newest refinement comment must hold them unedited; intake's refinement of this body is due and may supply them.melodic-automation-lanes commentedon Nov 4, 2026
Refinement. A read-only run explored this repository, searched its issues and pull requests, and read the allowed docs to draft what follows. On an issue a listed human filed, it counts as the issue's spec once posted, while unedited; it is still model text, so check it.
Intended outcome: pr.md describes /code-review:code-review as terminal-by-default and read-only without --comment, gates only the --comment form, and carries an updated as-of date and recheck trigger.
Related work
Open questions
RECOMMENDED: Yes if the command file confirms it. Then the Mutation gate covers only the --comment form, and a bare /code-review:code-review run on a PR is a permitted read-only pass. Basis: judgment
RECOMMENDED: List it as an optional read-only pass, not a default. The manual path stays the default, as pr.md treats the bundled command. Basis: plugins/review/skills/quality-gate/context/pr.md:50
RECOMMENDED: Yes. Replace the fired trigger with that one and update the date. Basis: plugins/review/skills/quality-gate/context/pr.md:36
RECOMMENDED: Yes, if they assert the plugin always posts. Narrow the eval at fanout/evals/evals.json:72 to the --comment form and check fanout/SKILL.md:166 for the same wording. Basis: plugins/review/skills/fanout/evals/evals.json:72
Sources read
To answer, post a comment that starts with this line:
Automated refinement by the lanes App. Answer with a comment whose first line is
**Decision recorded.**.kyle-sexton commentedon Nov 4, 2026
Decision recorded. Answers to the refinement's four questions, from the upstream command file read 2026-11-04.
Agent choice, not an owner answer.
Source:
plugins/code-review/commands/code-review.mdonanthropics/claude-plugins-officialmain, read directly withgh api(not through/discovery:research; it is the single primary source). Line numbers below areorigin/mainafter #12997.--commentargument was NOT provided, stop here. Do not post any GitHub comments." Only the--commentbranches post: step 7 posts a summary comment withgh pr commentwhen no issues were found, and step 9 posts inline comments. The command still listsBash(gh pr comment:*)and the inline-comment tool inallowed-tools, so the read-only default rests on its instructions, not a tool restriction;pr.mdshould say so. Step 1 also stops every run on a closed or draft PR, and when "Claude has already commented on this PR". A PR that already carries a review comment written by Claude, such as one from this repository's CI review lanes, can trip that stop, sopr.mdshould name it as a reason a default run may return nothing./code-review:code-reviewwithout--commentbeside the bundled/code-review <target>(pr.md:57-59); the manual path stays the default.pr.md:14-18, which already points at the README's Commands section) addscommands/code-review.md, with as-of 2026-11-04, and its recheck trigger becomes: the command file changes its--commenthandling, its step-1 stop conditions, or itsallowed-tools. The Mutation gate (pr.md:45-50) covers the plugin only in its--commentform.fanout/SKILL.md:166,fanout/context/findings-normalization.md:21and the expected output atfanout/evals/evals.json:69; the assertion atevals.json:72stays. SKILL.md:166 says "there is no mode that returns them to the session instead", findings-normalization.md:21 says the plugin "ends by posting its surviving findings as a PR comment", and evals.json:69 says it "always ends by commenting on the PR". Each is true only of the--commentform, so each is narrowed to it, and SKILL.md:166 says an opted-in dispatch passes--comment, since the comment fetch-back it describes needs a posted comment. Fanout still dispatches the plugin only on that opt-in: the default run's terminal output (step 7, "list each issue with a brief description") has no documented schema, and the parse row at findings-normalization.md:16 covers only the posted comment. So the no-dispatch-without-opt-in assertion at :72 stays true.Same plugin, may ride the same PR:
fanout/SKILL.md:166("parallel reviewers + confidence scorer") andfindings-normalization.md:16("0–100, filters <80") describe a scorer the current command no longer has; it validates findings with subagents (steps 5-6).plugins/review/skills/quality-gate/context/pr.mdno longer says this mode never runs/code-review:code-reviewas a read-only pass; it describes the plugin's default run as terminal output with no GitHub writes, says that default rests on the command's instructions rather than itsallowed-tools, and names the step-1 stops (a closed or draft PR, and a PR Claude has already commented on).pr.mdcovers/code-review:code-reviewonly with--comment, and the plugin's default run is listed as an optional read-only pass beside the bundled command.code-reviewplugin's pointer bullet inpr.mdpoints at the plugin README's Commands section and atcommands/code-review.md, its as-of date is 2026-11-04, and its recheck trigger is the one in answer 3.plugins/review/skills/fanout/SKILL.mdno longer says the plugin has no mode that returns findings to the session, and says an opted-in dispatch passes--comment.plugins/review/skills/fanout/context/findings-normalization.mdand the expected output of the post-to-PR case inplugins/review/skills/fanout/evals/evals.jsondescribe posting as the--commentform only.github-actions commentedon Nov 4, 2026
github-actions commentedon Nov 4, 2026