kin: auto-commit after pipeline

agents/runner.py

@@ -1177,6 +1177,38 @@ def _execute_department_head_step(
     }
 
 
+# ---------------------------------------------------------------------------
+# Watchdog helpers
+# ---------------------------------------------------------------------------
+
+import errno as _errno
+
+
+def _check_parent_alive(
+    conn: sqlite3.Connection,
+    pipeline: dict,
+    task_id: str,
+    project_id: str,
+) -> bool:
+    """Check if parent process is alive. Returns True if pipeline should abort.
+
+    Only treats ESRCH (no such process) as dead parent.
+    PermissionError (pid 1 / init) and ValueError are ignored — pipeline continues.
+    """
+    ppid = os.getppid()
+    try:
+        os.kill(ppid, 0)
+    except OSError as exc:
+        if exc.errno == _errno.ESRCH:
+            reason = f"Parent process died unexpectedly (PID {ppid})"
+            _logger.warning("Pipeline %s: %s — aborting", pipeline["id"], reason)
+            models.update_pipeline(conn, pipeline["id"], status="failed")
+            models.update_task(conn, task_id, status="blocked", blocked_reason=reason)
+            return True
+        # PermissionError (EPERM) — process exists but we can't signal it: continue
+    return False
+
+
 # ---------------------------------------------------------------------------
 # Pipeline executor
 # ---------------------------------------------------------------------------
@@ -1234,6 +1266,9 @@ def run_pipeline(
         pipeline = models.create_pipeline(
             conn, task_id, project_id, route_type, steps,
         )
+        # Save PID so watchdog can detect dead subprocesses (KIN-099)
+        models.update_pipeline(conn, pipeline["id"], pid=os.getpid())
+        pipeline["pid"] = os.getpid()
         models.update_task(conn, task_id, status="in_progress")
 
     results = []
@@ -1248,6 +1283,19 @@ def run_pipeline(
         model = step.get("model", "sonnet")
         brief = step.get("brief")
 
+        # Check parent process is still alive (KIN-099 watchdog)
+        if not dry_run and pipeline:
+            if _check_parent_alive(conn, pipeline, task_id, project_id):
+                return {
+                    "success": False,
+                    "error": "parent_process_died",
+                    "steps_completed": i,
+                    "total_cost": total_cost,
+                    "total_tokens": total_tokens,
+                    "total_duration": total_duration,
+                    "results": results,
+                }
+
         # Worktree isolation: opt-in per project, for write-capable roles
         _WORKTREE_ROLES = {"backend_dev", "frontend_dev", "debugger"}
         worktree_path = None

agents/specialists.yaml

@@ -332,3 +332,11 @@ routes:
   dept_marketing:
     steps: [marketing_head]
     description: "Marketing task routed through department head"
+
+  dept_infra:
+    steps: [infra_head]
+    description: "Infrastructure task routed through department head"
+
+  dept_research:
+    steps: [research_head]
+    description: "Research task routed through department head"