parallel-planner-with-review: a branch ahead of HEAD from an earlier cycle reaches the merger without review #1628

What happens

In the parallel-planner-with-review template, the reviewer only runs when this cycle's implementer made commits:

// src/templates/parallel-planner-with-review/main.mts, Phase 2
// Only review if the implementer produced commits
if (implement.commits.length > 0) {
  const review = await sandbox.run({ name: "reviewer", ... });

Since #1531, the merge phase no longer looks at this cycle's commits. It passes the merger every fulfilled branch that is ahead of HEAD (isAheadOfHead, i.e. git rev-list --count HEAD..<branch> > 0), whichever cycle made the commits.

Those two gates now disagree, and a branch can reach the merger without ever being reviewed:

1. Cycle N: the implementer commits on issue-42. The reviewer then fails (network error, sandbox crash, the agent erroring out), so the pipeline promise rejects and the branch is left out of the merge.
2. Cycle N+1: the planner picks issue 42 again, since it's still open. The implementer finds the work already on the branch and makes no new commits, so implement.commits.length === 0 and the reviewer is skipped. The pipeline fulfils.
3. The merge phase sees that issue-42 is ahead of HEAD and hands it to the merger, with no review.

The same happens when the implementer itself commits and then throws in cycle N: in that case its commits are never reviewed at all.

Where

• src/templates/parallel-planner-with-review/main.mts: the implement.commits.length > 0 gate in Phase 2, compared with the isAheadOfHead filter before Phase 3.
• src/templates/parallel-planner/main.mts has the same isAheadOfHead merge gate but no reviewer, so it isn't affected. sequential-reviewer creates a fresh branch every iteration and has no merge phase, so it isn't affected either.

Expected

A branch that is ahead of HEAD and hasn't passed review gets reviewed before it can merge. In practice, every branch the merger receives in a cycle has had a reviewer run complete on it in that same cycle.

Prior work

• fix: gate merge phase on "branch ahead of HEAD", not this cycle's commits #1531 moved the merge gate to "branch ahead of HEAD", which exposed this.
• fix: retry review for existing branch work #938 (closed) fixed the retry case before fix: gate merge phase on "branch ahead of HEAD", not this cycle's commits #1531. It re-ran the reviewer when the branch had unmerged commits (hasUnmergedCommits), and added a hasWork flag for the merge filter. fix: gate merge phase on "branch ahead of HEAD", not this cycle's commits #1531 has since made the hasWork half unnecessary, but the reviewer half still applies.

This issue was written by an AI agent (Claude Code).