feat: T-1.21 + T-1.22 + P1-23 — 打通 dispatch→Worker→merge 全链路

T-1.21 (P1-22): Dispatch → Worker 桥接
- eng_mode: spawn_workers() 为每个 ready 任务生成标准 Skill 调用参数
- commands/eng.md: dispatch 段从意图描述改为可执行5步伪代码
  Agent 不再需要猜测工具名、参数格式、task-text 来源

T-1.22 (P1-24): Merge → TaskGraph 状态同步
- merge_worker_result Phase 6.5: 写回 task-graph.json 节点 status
- merge 后节点状态 DONE，ready_tasks() 过滤已完成任务

_select_ready_tasks 修复：
- DAG ready 为空时不再 fallback 到 todo.md（todo.md 状态陈旧时导致重复派发）

P1-23: commands/eng.md dispatch 指令操作化
- 保留 Eng 工具白名单不变（极端接管需要 Write/Edit）
- Arc/Eng 约束非对称性是刻意的设计决策

63+4 项测试全通过，含端到端全链路测试

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

lib/air_runtime/modes/eng_mode.py

@@ -556,6 +556,16 @@ def merge_worker_result(project_root: Path, result_path: Path) -> dict:
         state["lastMergeAt"] = now_iso()
         atomic_json_write(paths["state"], state)
 
+        # Phase 6.5: 同步 task-graph.json 节点状态（P1-24）
+        tg_json = airplan_root(project_root) / "state" / "airarc" / "reviews" / "task-graph.json"
+        if tg_json.exists():
+            from air_runtime.modes.arc_mode import _export_task_graph_json
+            graph = TaskGraph.load(tg_json)
+            if task_id in graph.nodes:
+                new_status = "DONE" if status == "done" else status.upper()
+                graph.nodes[task_id].status = new_status
+                _export_task_graph_json(graph, tg_json)
+
     log.emit(MERGE_COMPLETED, {
         "taskId": task_id,
         "status": status,
@@ -586,17 +596,17 @@ def merge_worker_result(project_root: Path, result_path: Path) -> dict:
 
 
 def _select_ready_tasks(project_root: Path, max_count: int) -> list[str]:
-    """优先从 task-graph.json 的 DAG 计算 in-degree 为 0 的 TODO task，fallback parse_tasks。"""
+    """优先从 task-graph.json 的 DAG 计算 in-degree 为 0 的 TODO task。
+    DAG 中 ready 为空意味着无任务可派发（全部完成或全部被依赖阻塞），不应 fallback 到 todo.md。"""
     task_graph_json = airplan_root(project_root) / "state" / "airarc" / "reviews" / "task-graph.json"
     if task_graph_json.exists():
         try:
             graph = TaskGraph.load(task_graph_json)
             ready = graph.ready_tasks()
-            if ready:
-                return ready[:max_count]
+            return ready[:max_count]  # 空列表也是正确答案，不 fallback
         except Exception:
             pass
-    # fallback
+    # fallback：仅在 task-graph.json 不存在时使用 todo.md
     todo = get_todo_path(project_root)
     if not todo.exists():
         return []
@@ -604,6 +614,25 @@ def _select_ready_tasks(project_root: Path, max_count: int) -> list[str]:
     return [t.task_id for t in tasks if t.status == "TODO"][:max_count]
 
 
+def spawn_workers(project_root: Path, task_ids: list[str]) -> list[dict]:
+    """T-1.21: 为每个 ready 任务准备 spawn 指令，返回 Agent 可直接消费的 Skill 调用参数列表。
+    不实际启动子进程——启动由 Agent 框架的 Skill 工具完成。
+    """
+    tg_json = airplan_root(project_root) / "state" / "airarc" / "reviews" / "task-graph.json"
+    graph = TaskGraph.load(tg_json) if tg_json.exists() else TaskGraph()
+    instructions = []
+    for tid in task_ids:
+        node = graph.nodes.get(tid)
+        task_text = node.task if node else ""
+        instructions.append({
+            "skill": "airplan",
+            "args": f"do --sub enter --task-id {tid} --task-text '{task_text}' --project .",
+            "taskId": tid,
+            "taskText": task_text,
+        })
+    return instructions
+
+
 def handle_adr_invalidation(project_root: Path, adr_id: str) -> dict:
     """P1-21: ADR 变更级联失效处理。