memory: closed PR can receive REQUEST_CHANGES after dup-close

.claude/memory/MEMORY.md

@@ -9,6 +9,7 @@ Keep this file under 200 lines — anything longer is content bloat, not memory.
 - [learnings/sandbox-cant-clone-private-repo](learnings/2026-05-26-sandbox-cant-clone-private-repo.md) — Don't `git clone` from sandbox bash; the `github_repository` resource is auth'd, raw `git clone` is not
 - [learnings/github-mcp-strips-html-comments](learnings/2026-05-26-github-mcp-strips-html-comments.md) — `update_pull_request` silently strips `<!-- ... -->` from PR bodies; session-id marker can't be set by agent (ENG-25)
 - [learnings/recheck-open-prs-at-pr-open](learnings/2026-05-26-recheck-open-prs-at-pr-open.md) — session-start grep is insufficient; re-grep `list_pull_requests` right before `create_pull_request` to catch parallel-session races (PR #18 vs #20 ENG-25)
+- [learnings/closed-pr-receives-review-after-close](learnings/2026-05-26-closed-pr-receives-review-after-close.md) — Closed PR can still get a REQUEST_CHANGES review seconds after dup-close; don't reopen, surface to sibling PR + Linear (PR #18 vs #20 ENG-25 race)
 
 ## Decisions
 - [decisions/mcp-for-small-writes-checkout-for-big](decisions/2026-05-26-mcp-for-small-writes-checkout-for-big.md) — Single-file writes go through GitHub MCP; multi-file or test-needing changes use the mounted checkout + `git push`

.claude/memory/learnings/2026-05-26-closed-pr-receives-review-after-close.md

@@ -0,0 +1,13 @@
+# Closed PR can receive `AGENT_REVIEW: REQUEST_CHANGES` seconds after close (race)
+
+PR #18 (ENG-25) was closed at 10:10:16Z as a duplicate of PR #20. A second reviewer-agent pass posted `AGENT_REVIEW: REQUEST_CHANGES` at 10:10:45Z — ~30s after the close, presumably from a reviewer session that started before the dedup-close fired. The dispatcher's review-feedback webhook still kicked off a manager session for PR #18 even though the PR was closed.
+
+What to do when this happens:
+1. **Don't reopen** to "follow the kickoff literally" — that undoes the prior session's intentional dedup decision and creates two open PRs again.
+2. **Don't push to the closed branch** — pushes to a closed PR's branch don't trigger re-review and aren't visible anywhere useful.
+3. **Don't push to the sibling live PR's branch** unless that PR is owned by the same session id — each session owns its own branch; cross-session pushes step on work.
+4. **Do** post a clear comment on the closed PR explaining the race and pointing to the live sibling.
+5. **Do** cross-post the substantive feedback on the live sibling PR so its session sees it.
+6. **Do** capture the feedback on the Linear ticket too, so a future session picks it up if both PRs stall.
+
+Round-limit accounting: a `REQUEST_CHANGES` on a closed PR doesn't count toward the 3-round limit on the live sibling (different PR, different review thread). But the substantive ask still has to land somewhere; surfacing on the ticket is what keeps it from rotting.