huskies: merge 498_bug_stale_merge_job_lock_prevents_new_merges_after_agent_dies

server/src/agents/pool/pipeline/merge.rs

@@ -212,6 +212,70 @@ mod tests {
             .unwrap();
     }
 
+    // ── bug 498: stale Running job blocks retry ───────────────────────────────
+
+    /// Regression test for bug 498: a Running merge job left behind by a killed
+    /// mergemaster must not block the next call to start_merge_agent_work.
+    ///
+    /// Before the fix: start_merge_agent_work would return "Merge already in
+    /// progress" when a Running entry existed, even after the mergemaster died.
+    /// After the fix: the entry is cleared when the mergemaster exits, so a new
+    /// call succeeds.
+    #[tokio::test]
+    async fn stale_running_merge_job_is_cleared_and_retry_succeeds() {
+        use tempfile::tempdir;
+
+        let tmp = tempdir().unwrap();
+        let repo = tmp.path();
+        init_git_repo(repo);
+
+        let pool = Arc::new(AgentPool::new_test(3001));
+
+        // Inject a stale Running entry, simulating a mergemaster that died
+        // before the merge pipeline completed.
+        {
+            let mut jobs = pool.merge_jobs.lock().unwrap();
+            jobs.insert(
+                "77_story_stale".to_string(),
+                MergeJob {
+                    story_id: "77_story_stale".to_string(),
+                    status: MergeJobStatus::Running,
+                },
+            );
+        }
+
+        // With a stale Running entry, start_merge_agent_work must be blocked.
+        let blocked = pool.start_merge_agent_work(repo, "77_story_stale");
+        assert!(
+            blocked.is_err(),
+            "start_merge_agent_work must be blocked while Running job exists"
+        );
+        let err_msg = blocked.unwrap_err();
+        assert!(
+            err_msg.contains("already in progress"),
+            "unexpected error: {err_msg}"
+        );
+
+        // Simulate the mergemaster exit path: clear the stale Running entry.
+        {
+            let mut jobs = pool.merge_jobs.lock().unwrap();
+            if let Some(job) = jobs.get("77_story_stale")
+                && matches!(job.status, MergeJobStatus::Running)
+            {
+                jobs.remove("77_story_stale");
+            }
+        }
+
+        // After clearing, start_merge_agent_work must succeed (it will fail
+        // the pipeline because there's no feature branch, but it must not be
+        // blocked by "Merge already in progress").
+        let result = pool.start_merge_agent_work(repo, "77_story_stale");
+        assert!(
+            result.is_ok(),
+            "start_merge_agent_work must succeed after stale Running job is cleared; got: {result:?}"
+        );
+    }
+
     // ── merge_agent_work tests ────────────────────────────────────────────────
 
     /// Helper: start a merge and poll until terminal state.

server/src/agents/pool/start.rs

@@ -320,6 +320,7 @@ impl AgentPool {
         let log_writer_clone = log_writer.clone();
         let child_killers_clone = self.child_killers.clone();
         let watcher_tx_clone = self.watcher_tx.clone();
+        let merge_jobs_clone = Arc::clone(&self.merge_jobs);
 
         // Spawn the background task. Worktree creation and agent launch happen here
         // so `start_agent` returns immediately after registering the agent as
@@ -524,6 +525,15 @@ impl AgentPool {
                                 (tx_clone.clone(), result.session_id)
                             }
                         };
+                        // Clear any stale Running merge job so the next mergemaster
+                        // can call start_merge_agent_work without hitting "Merge
+                        // already in progress" (bug 498).
+                        if let Ok(mut jobs) = merge_jobs_clone.lock()
+                            && let Some(job) = jobs.get(&sid)
+                            && matches!(job.status, crate::agents::merge::MergeJobStatus::Running)
+                        {
+                            jobs.remove(&sid);
+                        }
                         let _ = tx_done.send(AgentEvent::Done {
                             story_id: sid.clone(),
                             agent_name: aname.clone(),