[getappz]

Redispatching fixes for a live PR now preserves metadata.pr and its review state, including when GitHub cannot be reached. keep_pr: true overrides the lookup. Merged, closed, or missing PRs are cleared together with review_threads and coderabbit_repair_announced_for in one metadata write. A replacement PR attached during the lookup is retained.

The closed-PR supervisor also binds its cleanup verdict to the PR number confirmed closed, including its no-assignee SQL fallback. A caller-level regression replaces PR #501 with #502 after the supervisor guard and verifies the replacement PR and review metadata survive in both assignee paths.

Depends on #946 for PR recovery. The redispatch handler and its new MCP test were split into modules to satisfy the repository's file-size gate.

Validation at bfa6231a:

• cargo fmt --check and cargo fmt --all --check: pass after cargo fmt fixed test formatting.
• Exact local workspace clippy, cargo test --workspace, and the new regression were run to completion; verification is blocked by disk exhaustion (OS error 112) in the shared Cargo target directory. Earlier attempts also waited on the shared build lock. No local test-pass claim is made for this revision.
• New-head CI formatting, exact workspace clippy, Linux/Windows workspace tests and doctests, Rust analysis, and CI Green pass. Linux reports 4,614 passed / 10 skipped and explicitly passes the new race regression. Windows completes successfully in 19m31s. Foreground gh pr checks 951 --watch --interval 20 completes with exit 0. CI run: https://github.com/getappz/agentflare/actions/runs/39732252509
• Initial implementation's backend redispatch tests: 12 passed. Its CI was green; current-head results are tracked separately above.

---

Opened by codex on i-flare-c997d745ae66 (flared:c997d745ae66) for item #736 via agentflare.

[coderabbitai]

Warning

Review limit reached

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

Next included review available in 29 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used all 2 included reviews currently available. Your 56 included PR review attempts over the past 7 days set your current allowance at 2 reviews per hour.

Learn how review limits work.

Review configuration:

Run configuration

• Source mode: Auto
• Repository: getappz/agentflare
• Branch: task/736-redispatch-clears-metadata-pr-even-when
• File: .coderabbit.yaml
• Commit: 11cad65
• Layers: Repository YAML
• Review profile: CHILL
• Plan: Essentials
• Run ID: 78297494-4cb6-4a79-8432-972d22a000bf

Commits

Reviewing files that changed from the base of the PR and between fcd7540 and 11cad65.

Files selected for processing: 17

• crates/agentflare-backend/src/item/claim.rs
• crates/agentflare-backend/src/item/tests.rs
• crates/agentflare-backend/src/item/tests_finalize.rs
• src/claim_liveness.rs
• src/mcp_server/handoff.rs
• src/mcp_server/item.rs
• src/mcp_server/item_redispatch.rs
• src/mcp_server/tests/action_tests.rs
• src/mcp_server/tests/redispatch_tests.rs
• src/mcp_server/types.rs
• src/supervisor.rs
• src/supervisor/merge.rs
• src/supervisor/tests/multi_project_sweep_tests.rs
• src/supervisor/tests/stray_pr_tests.rs
• src/supervisor_project_scope_tests.rs
• src/worktree/mod.rs
• src/worktree/mod_tests.rs

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}

[getappz]

Heads-up on one interaction with a change that landed after this PR was opened: #953 (item done now completes an item whose PR has already merged, and metadata.pr survives that completion).

Two things worth knowing, both in your favour:

1. This PR removes the first line of defence chore: item done leaves an item stuck In Progress when its PR was merged externally before finalize ran (seen on #283 / PR #910) #953 relied on. chore: item done leaves an item stuck In Progress when its PR was merged externally before finalize ran (seen on #283 / PR #910) #953's argument was "it keys on metadata.pr, and redispatch clears that — so an item re-dispatched for more work carries no PR reference and can never be completed by someone else's merge". Under this PR's PrRetention::Keep, a redispatched item with a live PR keeps metadata.pr, so that argument no longer holds.

2. The load-bearing condition is unaffected. chore: item done leaves an item stuck In Progress when its PR was merged externally before finalize ran (seen on #283 / PR #910) #953 also requires the merge to have covered the branch — the local branch tip has to be exactly the PR's head sha (worktree::branch_is_covered_by_merged_pr). A retried attempt that pushes to the same branch moves the tip, so a later merge of that PR completes the item exactly when it should; and a redispatched item whose branch has moved on past the merged commit is not completed away. The one narrow case that changes is "redispatch → the agent commits nothing → the old PR merges", where the item now completes instead of staying open — which reads as correct anyway (nothing was delivered, and the work did merge).

Net: the two changes agree. PrRetention::Keep + the branch-coverage check compose into "the PR is the unit of work", which is the same shape as PrRetention::Clear for the dead-PR case.

One note on ordering: PrRetention::Clear covers "the PR is merged" — and with #953 a merged PR can now also arrive at done. Both paths want the same outcome (the item is done), so no conflict there either.

Not a blocker — just wanted the reasoning on the record, since #953's doc comment cites the old redispatch behaviour and will need a line updated when this lands.