huskies: merge 1178 bug MergeFailureFinal is a trap state: successful re-merge cannot mark the story done

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

@@ -361,15 +361,23 @@ impl AgentPool {
         }
 
         let story_archived = crate::agents::lifecycle::move_story_to_done(story_id).is_ok();
-        if story_archived {
-            self.remove_agents_for_story(story_id).await;
-        }
 
-        let worktree_cleaned_up = if wt_path.exists() {
-            let config = crate::config::ProjectConfig::load(project_root).unwrap_or_default();
-            worktree::remove_worktree_by_story_id(project_root, story_id, &config)
-                .await
-                .is_ok()
+        // Story 1178: only delete the feature branch once the state transition
+        // to Done is confirmed. Deleting it unconditionally here meant a
+        // successful squash merge whose CRDT transition failed (e.g. the story
+        // was in a stage `move_story_to_done` didn't yet handle) would still
+        // lose its feature branch, making the failure unrecoverable — the
+        // story couldn't be retried because the branch it needed was gone.
+        let worktree_cleaned_up = if story_archived {
+            self.remove_agents_for_story(story_id).await;
+            if wt_path.exists() {
+                let config = crate::config::ProjectConfig::load(project_root).unwrap_or_default();
+                worktree::remove_worktree_by_story_id(project_root, story_id, &config)
+                    .await
+                    .is_ok()
+            } else {
+                false
+            }
         } else {
             false
         };

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

@@ -409,6 +409,97 @@ async fn merge_agent_work_succeeds_on_clean_branch() {
     }
 }
 
+/// Regression test (story 1178, AC3): when the squash merge itself succeeds
+/// but the CRDT state transition to Done fails (here: no CRDT entry exists
+/// for the story, so `move_story_to_done` errors and `story_archived` is
+/// false), the feature branch must NOT be deleted — otherwise a retry has
+/// nothing to merge from.
+#[tokio::test]
+async fn merge_success_without_story_archived_keeps_feature_branch() {
+    let _serial = serial_test_lock();
+    use std::fs;
+    use tempfile::tempdir;
+
+    crate::crdt_state::init_for_test();
+    let tmp = tempdir().unwrap();
+    let repo = tmp.path();
+    init_git_repo(repo);
+
+    let branch = "feature/story-1178_branch_kept";
+    Command::new("git")
+        .args(["checkout", "-b", branch])
+        .current_dir(repo)
+        .output()
+        .unwrap();
+    fs::write(repo.join("feature.txt"), "feature content").unwrap();
+    Command::new("git")
+        .args(["add", "."])
+        .current_dir(repo)
+        .output()
+        .unwrap();
+    Command::new("git")
+        .args(["commit", "-m", "add feature"])
+        .current_dir(repo)
+        .output()
+        .unwrap();
+    Command::new("git")
+        .args(["checkout", "master"])
+        .current_dir(repo)
+        .output()
+        .unwrap();
+
+    let merge_dir = repo.join(".huskies/work/4_merge");
+    fs::create_dir_all(&merge_dir).unwrap();
+    fs::write(
+        merge_dir.join("1178_branch_kept.md"),
+        "---\nname: Branch Kept Test\n---\n",
+    )
+    .unwrap();
+    Command::new("git")
+        .args(["add", "."])
+        .current_dir(repo)
+        .output()
+        .unwrap();
+    Command::new("git")
+        .args(["commit", "-m", "add story in merge"])
+        .current_dir(repo)
+        .output()
+        .unwrap();
+
+    let pool = Arc::new(AgentPool::new_test(3001));
+    // Note: no CRDT entry is written for this story, so `move_story_to_done`
+    // will fail with NotFound — `story_archived` will be false regardless of
+    // git merge outcome. That is exactly the scenario this test protects.
+    let job = run_merge_to_completion(&pool, repo, "1178_branch_kept").await;
+
+    let MergeJobStatus::Completed(report) = &job.status else {
+        panic!("expected a completed job, got: {:?}", job.status);
+    };
+    if matches!(
+        report.result,
+        crate::agents::merge::MergeResult::Success { .. }
+    ) {
+        assert!(
+            !report.story_archived,
+            "story_archived should be false: no CRDT entry exists for this story"
+        );
+        assert!(
+            !report.worktree_cleaned_up,
+            "worktree/branch must not be cleaned up when story_archived is false"
+        );
+        let branch_check = Command::new("git")
+            .args(["rev-parse", "--verify", branch])
+            .current_dir(repo)
+            .output()
+            .unwrap();
+        assert!(
+            branch_check.status.success(),
+            "feature branch '{branch}' must still exist after a merge whose state \
+             transition to Done failed, so a retry stays possible"
+        );
+    }
+}
+
 // ── quality gate ordering test ────────────────────────────────
 
 /// Regression test for bug 142: quality gates must run BEFORE the fast-forward