review-fix
Skill48 starsChanged 3 days agoApache-2.0View on GitHub
Use when about to act on PR review comments on an active branch — before you start editing — or when asked to address review feedback, clean up a PR's commit story, or update a PR after a review round.
- Deletes or force-pushes
- Commits and pushes
11 other projects use a version of this.
Adopted by
Most recently taken up by
- davidvornholt/mail-mcptheir own version, first seen Nov 15, 2026
- davidvornholt/mailboxtheir own version, first seen Nov 15, 2026
- nwiizo/ccswarmtheir own version, first seen Nov 6, 2026
- raine/skillstheir own version, first seen Nov 1, 2026
- 200iqlabs/shared-skillstheir own version, first seen Oct 31, 2026
- pcode-xyz/simple-skillstheir own version, first seen Oct 20, 2026
Used with
Projects with it also use review.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
SentrySkillreview
Fixture review skill.
Also in 5 projects that use this
Other versions11 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 11
- nwiizo/ccswarm★ 156same name
- DasDigitaleMomentum/opencode-processing-skills★ 60same name
- jhd3197/CachiBot★ 19same name
- EternallLight/tgc-skills★ 11same name
- 200iqlabs/shared-skills★ 4same name
- pcode-xyz/simple-skills★ 4same name
- davidvornholt/mail-mcp★ 2same name
- davidvornholt/mailbox★ 2same name
- raine/skills★ 2same name
- marcuskrogh/skills★ 1same name
1 more not listed
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 repository11 other files it gives its agents.
- AGENTS.mdAGENTS.md
- amend-adrSkill
- code-review-and-qualitySkill
- credential-incidentSkill
- file-issueSkill
- prove-event-workflowSkill
- releaseSkill
- review-prSkill
- security-and-hardeningSkill
- security-reviewSkill
- visual-pr-shotsSkill
The file
We split this skill into the rules it gives an agent
2 rules. 0 are common, 1 is distinctive, and 1 is very specific to this project.
0
Common
Most of the 10 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 83
Always
--force-with-lease,: never bare --force — it aborts if the remote moved under you. - Distinctive: line 84
Inline review comments go "outdated": after a history rewrite, because the commits they were anchored to no longer exist.
“Common” is measured: kept by at least 6 in 10 of the 10 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
/review-fix — fold review feedback into the original commits
The authority for what to do with review feedback is docs/CONVENTIONS.md (§ Workflow, ADR-0016): fold review-driven fixes into the commit that introduced the issue; do not add follow-up "fix review comments" commits; rewrite branch history and update the same PR; keep a separate commit only when the review introduced genuinely new scope. This skill is the procedure for applying that rule safely — do not restate the rule, follow it.
Why this exists
The recurring failure is reaching for follow-up "address review" commits because they feel safer than rewriting history. They are not — they litter the PR's story with corrections to commits that should simply have been right. The rewrite below is mechanical and provable; this skill removes the excuse.
Procedure
- Read the whole review. Inline comments live in a different place than the summary:
gh pr view <n> --json reviews --jq '.reviews[] | "\(.author.login) [\(.state)]\n\(.body)\n"'gh api repos/<owner>/<repo>/pulls/<n>/comments --jq '.[] | "\(.path):\(.line // .original_line)\n\(.body)\n[hunk]\n\(.diff_hunk)\n"'gh api repos/<owner>/<repo>/issues/<n>/commentsfor PR-level (non-inline) comments.
- Classify each comment — after checking it is right. Reviewers are sometimes wrong: verify the comment's claim against the code first, and reply on the PR with your reasoning instead of implementing feedback that is incorrect. A comment phrased as a question gets a serious answer on the PR — do not read it as rhetorical unless that is unmistakable; it may need no code change at all. For the comments that hold: correction to existing code → fold into the commit that introduced it; genuinely new scope → its own new commit (allowed by the policy). If a comment's intent is ambiguous (e.g. "this is unnecessary" on a method — the whole thing, or one line?), ask before acting; the wrong read can delete coverage.
- Map each fix to its origin commit. For each fix, find the commit that introduced the lines
it touches:
git log --oneline <base>..HEAD -- <path>andgit log -S '<symbol>' <base>..HEAD, where<base>is$(git merge-base main HEAD). One fix can span several origin commits (a rename left half-done across a refactor commit and the feature commits built on it) — each part folds into its own origin commit. A confirmed issue is also rarely unique to the flagged line: sweep the branch diff for other instances of the same mistake and map each one as its own fix. - Apply each fix as a fixup commit. For each fix-part: make the edit, then
git add <files> && git commit --fixup=<origin-sha>. A fix that spans several origin commits is applied part by part — edit part A, fixup onto its origin, edit part B, fixup onto its origin. New-scope commits (step 2) are also made now, as ordinary commits on top — the rebase leaves them in place. When everything is in, the tree already equals the final result: run the fast lanes once now (make lint && make test && make itest). Then snapshot:git branch backup/<name>. - Fold everything with one rebase.
git rebase --autosquash $(git merge-base main HEAD)— since git 2.44 this works non-interactively, so the blocked-iflag is not needed. Original commit messages are preserved automatically. On a conflict, resolve toward what the commit being replayed should itself contain on top of the already-rebuilt commits — never by pasting in the final tree, which would smuggle later changes into earlier commits — thengit add <files>andGIT_EDITOR=true git rebase --continue. A wrong resolution cannot slip through: step 6 catches it. - Prove it.
git diff backup/<name> --statmust be empty — the tree is byte-identical, only redistributed across the right commits — andgit log --oneline <base>..HEADreads as the original story with no "fix review" commits and no leftoverfixup!subjects (git before 2.44 silently ignores--autosquashin a non-interactive rebase, so a leftover means the fold never ran). Do not re-run the fast lanes: an identical tree cannot produce different results, and step 4 already tested it. If the diff is not empty, recover withgit reset --hard backup/<name>and retry (mid-rebase,git rebase --abortreturns there too). - Update the same PR.
git push --force-with-lease. The PR head moves automatically; the title, number, and base are untouched.
GitHub gotchas (learned the hard way)
- Never rename the PR's branch to fix a stale name. GitHub's branch-rename closes the PR and it cannot be reopened against the vanished branch. A stale branch name is harmless (it is not part of merged history); leave it, or close the PR and open a fresh one — do not rename in place.
- Always
--force-with-lease, never bare--force— it aborts if the remote moved under you. - Inline review comments go "outdated" after a history rewrite, because the commits they were anchored to no longer exist. That is the accepted cost of the policy; the comments stay readable in the conversation. Optionally reply to each noting the commit that now carries the fix.
- Take the
backup/<name>ref before the rebase, always. It is the proof anchor in step 6 and the recovery path if anything goes wrong.
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.

