← ClaudeAtlas

rereviewing-a-github-pr-after-replylisted

Re-review a GitHub PR after the author replies or pushes: list what others said and pushed since my last review, check every commit they claim fixed things is actually on the head (force-push and rebase can drop it), range-diff old vs new head, re-check their factual claims in code, then post a review that supersedes my old verdict. Triggers on: re-review, look at the PR's updates, author says fixed, they addressed my comments, is the fix on the branch, did the force-push drop the commit, flip my request changes, check the one I already reviewed.
sblattj/agent-chandlery · ★ 0 · Code & Development · score 72
Install: claude install-skill sblattj/agent-chandlery
## More triggers pr-rereview-after-reply (old name) # Re-review a PR after the author's reply Answers: since my last review, what did they say, what actually changed on the head, and does my verdict still hold. Works on any GitHub repo: the PR URL carries `owner/repo`. ## Do this ```bash REPO_DIR=<local clone> ${CLAUDE_SKILL_DIR:-${SKILL_DIR:-~/.claude/skills/rereviewing-a-github-pr-after-reply}}/rereview-status.sh <pr-url> [my-login] BACK=2 ... # measure from my 2nd-to-last review (e.g. after I already posted a newer one) ``` Read-only. Prints: my last review (state, commit), `old...head` compare status, every comment / inline reply / review / force-push by others since then, each commit SHA named in their replies marked `ON HEAD` or `NOT ON HEAD`, and a `git range-diff` of old vs new head when `REPO_DIR` is set. `my-login` defaults to the account `gh api user` returns. Then: 1. **Any `NOT ON HEAD`**, or a range-diff with only `=` rows while they say "fixed": the fix is not on the PR. Say so with the compare status and one at-head proof (a line that still has the old code), without naming who dropped it (see below). 2. Verify each factual claim in their replies in code at the head (`git show "${H}:path"`), then post one new review at the new head that supersedes the old state (APPROVE clears my CHANGES_REQUESTED). ## Why it's built this way (friction mined) - A reply said "the fixes are in `<sha>`", then a later force-push (a rebase onto the base, by a *different*