engineering-review-comments — Claude Code skill: what it does, install
engineering-review-comments turns rough review notes into PR comments with a severity label, an engineering reason and a concrete ask.
MIT 125 installs ★ 635 updated 2026-09-29
npx skills add agentara/skills --skill engineering-review-comments -a claude-codeWhat engineering-review-comments does
Feed the skill your rough review notes and it hands back a comment tagged by weight. The tags are fixed: BLOCKER, IMPORTANT, QUESTION, NIT, OPTIONAL or CONSIDER, FYI. Longer remarks get a three-part shape: what the code does, why it matters, what the author should change. Tiny ones stay tiny.
Ready-made phrasing samples are included, along with a procedure for pushback. Reread the author's argument first, then either concede and resolve, or restate their point and give the code-health reason. A section on frequent slips covers unlabelled comments that authors read as mandatory and teaching notes parked as blockers. The material is adapted from Google Engineering Practices.
Triggers are about wording. Ask it to phrase a comment and it works; a diff review, a PR description or a reply to the author is outside what it activates on.
What it helps with
- rewrite a blunt review note so it targets the code, not the person
- assign a severity label before posting a comment
- explain to the author why a change is required
- answer pushback in a thread without starting an argument
- split a finding into a reason and a concrete next step
Example prompts for the agent
- Turn my note “race condition here” into a review comment with a severity label and a reason.
- How should I word this naming remark so it doesn't read as a blocker?
- The author pushed back on my
BLOCKER— help me reply without dragging the thread out.
How to install engineering-review-comments
1Skills CLI (Vercel)
npx skills add agentara/skills --skill engineering-review-comments -a claude-code2Manually: put the files in place
~/.claude/skills/engineering-review-comments/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/engineering-review-comments/
1Skills CLI (Vercel)
npx skills add agentara/skills --skill engineering-review-comments -a codex2Manually: put the files in place
.agents/skills/engineering-review-comments/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add agentara/skills --skill engineering-review-comments -a cursor2Manually: put the files in place
.cursor/skills/engineering-review-comments/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install agentara/skills skills/engineering/engineering-review-comments2Manually: put the files in place
.github/skills/engineering-review-comments/SKILL.md1Skills CLI (Vercel)
npx skills add agentara/skills --skill engineering-review-comments -a gemini-cli2Platform CLI
gemini skills install https://github.com/agentara/skills --path skills/engineering/engineering-review-commentsManually: put the files in place
.devin/skills/engineering-review-comments/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add agentara/skills --skill engineering-review-comments -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about engineering-review-comments
How do I install engineering-review-comments in Claude Code?
Run npx skills add agentara/skills --skill engineering-review-comments -a claude-code. The skill file lands at ~/.claude/skills/engineering-review-comments/SKILL.md.
What does engineering-review-comments do?
It drafts and rewrites code review comments: picks a severity label, states the engineering reason and asks for a specific fix.
Does it run a full code review?
No. It works on the wording of comments rather than on the diff. PR descriptions and author replies are only touched when you ask for the text.
Source text
SKILL.md · MIT
--- name: engineering-review-comments description: > Use when drafting, rewriting, classifying, or improving code review comments as a reviewer. TRIGGER on rough review notes, requests to make comments clearer or kinder, severity labels, "how should I phrase this review comment?", "write review comments", author pushback, or turning findings into PR comments. If role is unclear, ask reviewer or author. DO NOT TRIGGER for full code review, PR descriptions, or author responses unless wording is requested. --- # Engineering Review Comments Review comments should improve the code and the developer's future judgment. They should be clear about severity, grounded in engineering reasoning, and directed at the code rather than the person. Adapted from Google Engineering Practices Documentation, especially "How to write code review comments" and "Handling pushback in code reviews." Source: <https://google.github.io/eng-practices/review/reviewer/comments.html>. License: CC-BY 3.0. ## Comment Shape Use this structure for non-trivial comments: ```text SEVERITY: Observation about the code. Why this matters: engineering reason, user impact, maintainability impact, or violated project rule. Requested action: concrete fix, question, or decision the author needs to make. ``` Keep tiny comments tiny. Use the full shape when intent could be misunderstood. ## Severity Labels - `BLOCKER`: must change before approval because correctness, maintainability, security, data, test, or operational risk is unacceptable. - `IMPORTANT`: should change before approval; can be waived only with an explicit rationale. - `QUESTION`: reviewer needs information before deciding. - `NIT`: minor polish. Do not block approval. - `OPTIONAL` or `CONSIDER`: a suggestion the author can decline. - `FYI`: educational context or future consideration; no action expected. ## Writing Rules - Comment on the code, design, tests, or behavior; do not characterize the author. - Explain why when asking for anything beyond obvious style or mechanical cleanup. - Prefer "this makes X harder because..." over "I don't like X." - Ask for simpler code when a review-thread explanation is needed to understand ordinary behavior. - Accept a code comment only when the code cannot reasonably express the why by itself. - Provide direct code or a specific design only when it is faster or safer than making the author infer it. - Otherwise describe the problem and let the author choose the implementation. - Point out good engineering choices when useful, and say why they are good. ## Wording Patterns Use these patterns directly when helpful: ```text BLOCKER: This path can return success before the write is durable. If the process crashes after this point, callers will believe the operation completed even though the data may be missing. Please either move the success response after the durable write or make the operation explicitly asynchronous and observable. ``` ```text QUESTION: What guarantees that this callback cannot run concurrently? If it can run in parallel, `pending` can be decremented twice from the same value. If the framework serializes this callback, please add a short comment at the boundary where that guarantee comes from. ``` ```text NIT: This helper name reads as if it validates the payload, but it only normalizes whitespace. Could we rename it to `normalizePayloadText`? ``` ```text OPTIONAL: Consider extracting this branch into a named helper. The current version is correct, but a helper would make the three policy cases easier to scan. ``` ## Handling Pushback When the author disagrees: 1. Re-evaluate their argument first. They may be closer to the code or have context you missed. 2. If they are right, say so plainly and resolve the comment. 3. If you still disagree, restate their point, then explain the code-health reason for your request. 4. Distinguish mandatory requests from preferences. 5. If the thread stops making progress, move to a higher-bandwidth discussion and summarize the decision back in the review. Do not keep arguing because you wrote the first comment. The goal is the best code, not defending the initial wording. ## Common Mistakes - Leaving unlabeled comments that authors interpret as mandatory. - Writing a correct critique in a tone that makes collaboration harder. - Asking the author to explain confusing code only in the review tool. - Turning every comment into a complete design assignment. - Treating educational notes as blockers.
© agentara · MIT · agentara/skills