Code review
Run one evidence-focused, risk-scaled official code review on a pull request and post the result. Only invoke when explicitly requested as `$code-review` or `/code-review`, or when `ship` selects it as the sole official-review fallback.From its SKILL.md
npx -y skills add comment-hq/skills --skill code-reviewAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
3.8 KB, 738 tokens by cl100k_base, as published. Nobody here has run it
code-review — one official review for the exact PR head
This is the posted official review path. It is distinct from review-loop,
which certifies bounded in-flight deltas. Do not run both local Codex review and
this skill for the same candidate.
Eligibility
Read the PR's state, base, exact head SHA, changed files, relevant
AGENTS.md/CLAUDE.md, and prior reviews. Stop for a closed/draft PR unless
ship deliberately opened a draft because this is the only available official
reviewer. Do not skip merely because an earlier head was reviewed: official
evidence must match the current head.
Risk-scaled review
Use one strong primary reviewer for correctness, regressions, missed requirements, and repository-instruction violations. Add one independent specialist lens only when the diff touches authorization/security, migrations/storage, protocol/compatibility, concurrency, native code, destructive behavior, or credible data-loss risk. Three reviewers are reserved for exceptional blast radius.
Lock the review to the PR's declared subject, acceptance criteria, invariants, and exact base/head delta. Read only the surrounding code and history needed to verify impact; untouched context is evidence, not a broader audit surface. A finding is in scope only if the PR delta introduced or changed it, or the delta makes an existing path violate an explicit requirement/invariant. For a lift promotion, consume the review receipt ledger and focus on receipt coverage, cross-slice composition, main-sync/conflict delta, migration/cut-over ordering, and uncovered commits instead of blindly repeating each certified slice.
Findings
Keep an issue only when evidence demonstrates a concrete correctness, security/privacy, data-loss, compatibility, migration, protocol, or stated- requirement failure on changed code. Verify debatable claims from code or a focused case. Exclude:
- style or preference;
- speculative refactors/general quality advice;
- unrelated pre-existing debt;
- unsupported hypotheticals;
- compiler/linter/type errors already covered by a required gate, unless the review shows that gate is missing;
- intentional behavior required by the task.
Prefer the simplest implementation that works for the current user need. Do not block shipping on speculative enterprise hardening, abstraction, or unlikely edge cases without a credible reachable failure. This does not relax concrete security/privacy, data-loss, migration, protocol, or correctness requirements.
Report a genuinely unrelated defect separately as an out-of-scope discovery; do not put it in the PR review or expand the current diff. The orchestrating agent searches for or files one focused GitHub issue and continues. If it is actively release-breaking or makes the current delivery unsafe, the orchestrator stops and asks the owner human to start a separate worktree job.
The main agent deduplicates and verifies the complete finding batch. Post one brief official review for the exact head SHA, citing each real issue. If none remain, post “No actionable issues found” and name the scope/lenses reviewed. Never ask a second panel to sample until silence.
After a fix
The edited PR is a new head. Review only the fix delta plus impacted invariants when the platform supports it, but make the posted result explicitly apply to the new exact head. Normal work gets at most two finding-bearing rounds before redesign, targeted proof, residual-risk recording, or human escalation.
Do not run builds or full tests inside this skill; ship owns candidate
evidence. Use gh for PR inspection and posting.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.
Gives 0 of the 12 instructions most code review skills give in 738 tokens
Counted across 668 of the 814 authors here whose files we hold, read 2026-09-06
- Provide technical reasoning when pushing backin 84 of 668, across 70 files
- Fix critical issues immediatelyin 77 of 668, across 60 files
- Dispatch a code reviewer subagentin 76 of 668, across 59 files
- Fix important issues before proceedingin 73 of 668, across 56 files
- Ask for clarification on unclear itemsin 68 of 668, across 56 files
- Verify feedback against codebase before implementationin 66 of 668, across 55 files
- Implement fixes one at a timein 64 of 668, across 53 files
- Test each fix individuallyin 62 of 668, across 51 files
- Restate technical requirements in own wordsin 57 of 668, across 46 files
- Reply to inline comments in the specific threadin 51 of 668, across 40 files
- Note minor issues for laterin 49 of 668, across 34 files
- Group findings by severityin 48 of 668, across 47 files
Said here and by no other author read
- Read repository guidance before evaluating findings
- Invoke review-judgment for classification and thread resolution
- Lock the panel to the specific review target
- Assess impact only within surrounding context
- Map configuration changes to an authoritative owner
- Review deltas against the specific Bead acceptance surface
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.