address-review-comments
Skill2 starsChanged 30 days agoMITView on GitHub
Analyze, decide on, challenge, and implement a PR's open review comments end to end, with a human in the loop. Use when addressing PR review feedback, responding to reviewer comments, or when the user says "address review comments".
4 other projects use a version of this.
Adopted by
Most recently taken up by
- javierbrea/eslint-plugin-boundariestheir own version, first seen Nov 16, 2026
- Hakktivist/claude-skillstheir own version, first seen Nov 2, 2026
- AlexisBalayre/claude-code-power-configword for word, first seen Oct 17, 2026
- tstapler/dotfilestheir own version, first seen Oct 10, 2026
Other versions4 projects use a version of this.
Word for word · 1
- AlexisBalayre/claude-code-power-config★ 2100% same
Edited · 0
None found.
Wrote their own version · 3
- javierbrea/eslint-plugin-boundaries★ 1ksame name
- tstapler/dotfiles★ 8same name
- Hakktivist/claude-skills★ 2same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository36 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- adapt-to-projectSkill
- backfill-issuesSkill
- cavemanSkill
- codebase-designSkill
- daily-noteSkill
- diagnosing-bugsSkill
- domain-modelingSkill
- find-dead-codeSkill
- fix-sonarSkill
- fix-wizSkill
- grillingSkill
- grill-meSkill
- grill-with-docsSkill
- handoffSkill
- implementSkill
- implement-specSkill
- improve-codebase-architectureSkill
- obsidian-vaultSkill
- pr-ci-reviewSkill
- pr-descriptionSkill
- prototypeSkill
- researchSkill
- retroSkill
- review-retroSkill
- tddSkill
- to-questionnaireSkill
- to-specSkill
- to-ticketsSkill
- wait-whatSkill
- wayfinderSkill
- wizardSkill
- wizSkill
- writing-for-agentsSkill
- zoom-outSkill
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 3 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 31
No code is written in this phase.
- Distinctive: line 43
Ask the user whether to commit/push and whether to reply to the threads.
- Distinctive: line 53
Only change what each comment asks for.
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Address review comments
Take the open review comments on a PR from "what is the reviewer really asking" through to an implemented, replied-to resolution. You steer; the human decides every judgment call. Handles every open thread, human or bot. Closed and resolved threads are context only.
Process
1. Gather
Resolve the PR (the argument, or the current branch's PR via gh pr view). Pull its open review threads through gh / the GraphQL API: each comment with its file, line, author, body, and thread state. Resolved threads are context, not work. Always read the actual current code, not your recollection of it; the user may have changed it.
2. Triage
Sort each open thread into:
- settled — a pure mechanical fix, or the author already agreed an approach in-thread. Goes straight to implementation.
- needs a decision — everything else.
3. Decide (no code yet)
For each needs a decision thread, investigate the reviewer's underlying concern, not just the symptom they pointed at, until you can lay out the realistic options with their trade-offs. Then route by what is actually in dispute:
- Correctness dispute ("this finding is wrong / a false positive"): spawn a validator subagent that adversarially re-checks the finding against the actual code, given the user's rebuttal. One round only. It returns concede (a false positive, with a code-grounded reason) or defend (the issue holds, with the concrete evidence). Bias to concede when it cannot defend with evidence: the user raised the doubt, so the burden is on the finding.
- Design or approach decision: route through the
/grillingskill rather than settling it unilaterally; always provide your recommended answer. If the comment opens an architectural, ADR-worthy decision (hard to reverse, or it shifts a documented term or boundary), run/domain-modelingalongside, so the resolution lands in the repo's vocabulary or an ADR.
No code is written in this phase. The goal is alignment. The validator and the grilling session inform the decision; the user is the final authority on every thread.
4. Checkpoint
If anything went through step 3, present a per-thread summary — the concern, the options considered, the decision reached, and the concrete change or an explicit "leave as-is" with rationale — and get the user's sign-off before implementing. When every thread was already settled, skip straight to implementation.
5. Implement
Apply the agreed changes. Each fix targets only what its comment raised: no opportunistic cleanup, no unrelated refactors, no removing ticket-referenced TODOs. Note any tangential improvements separately in the summary and ask before touching them.
6. Reply and resolve
Ask the user whether to commit/push and whether to reply to the threads. Then, per thread, post a single reply and set its state:
- Addressed by a change → reply with the rationale and "Addressed in
<commit>."; resolve the thread (GraphQLresolveReviewThread). - Agreed, no change needed / confirmed correct → a thumbs-up or short acknowledgement; resolve.
- Conceded challenge (the finding was a false positive) → reply that it was retracted as a false positive and why; resolve.
- Leave-as-is (intentional, out-of-scope, or a finding the user overrides after it was defended) → reply with the rationale but leave the thread open. Never unilaterally close a disagreement.
Invariants
- The user may change the code at any point; work from, and commit, the actual current state, not your recollection.
- Only change what each comment asks for.
- The human is the final authority on every thread; validators and grilling sessions inform, they do not decide.
- A conceded challenge and a leave-as-is pushback are both signals the review retro will later read. Reply and (for concessions) resolve them honestly; do not scrub the disagreement record.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

