wf-qa-gate

WF: QA Gate

Gate Check

cat .claude/.workflow-state/wave-<currentWave>-spawned.json

Determine currentWave from setup-complete.json. If spawn stamp missing: STOP — "Wave not spawned. Run Step 4a (wf-spawn) first."

Load agent IDs, task list, TICKET, TEAM_LEADER_NAME, workPrefix, worktreeDir, featureBranch from state files.

---

Your Task

Wait for coder completions, coordinate QA, and merge each passing task. Do not move to the next wave until ALL tasks in this wave are merged.

---

Checklist

1. Wait for Coder Completions

Wait for each coder-task-<N> to SendMessage with completion. When a coder reports done:

1. Note which task completed and which files changed
2. Notify QA: SendMessage(to: "qa-task-<N>", message: "Code ready. Files: <list>.")

RECOVERY: If a coder goes silent for >5 minutes, send one nudge. If no response after 2 minutes, check TaskOutput. If crashed, spawn a replacement on the same worktree with the same task file.

2. Handle QA Verdict

On PASS:

1. Emit: /claude-workflow:track qa.passed "Task #<N>" --task <N>
2. Rebase and merge:

   git -C <worktreePath> rebase <featureBranch>
   git checkout <featureBranch>
   git merge --no-ff <workbranch> -m "Merge <taskSlug>: <summary>"
   

3. Emit: /claude-workflow:track branch.merged "<taskSlug> to feature"
4. Cleanup:

   git worktree remove <worktreePath>
   git branch -d <workbranch>
   

5. Shut down both agents:

   SendMessage(to: "coder-task-<N>", message: {"type": "shutdown_request"})
   SendMessage(to: "qa-task-<N>", message: {"type": "shutdown_request"})
   

6. Update task file: status → "complete"

On FAIL (round < 3):

1. Emit: /claude-workflow:track qa.failed "Task #<N>, round <R>" --task <N>
2. Forward to coder: SendMessage(to: "coder-task-<N>", message: "QA FAIL round <R>. Issues: <list>")
3. Wait for coder to fix and message back
4. Shut down previous QA agent
5. Spawn NEW QA: name: "qa-task-<N>-r<round>" with same template
6. Wait for new verdict. Increment round. Repeat from step 1.