feat: agent definition v2 (registry, persistence, developerInstructions channel)

Registry: POST /tasks accepts agent_name, loading <name>.json
from repo-root agents/ (traversal-guarded, skill-pack re-validated);
agent and agent_name together is a 400. Persistence: tasks gains
agent_name/agent_version/agent_instructions (migrate + PG schema),
returned by GET /tasks; stored prompt stays clean. Channel: local
and docker dispatch pass --developer-instructions to host.py
dispatch -> create_thread(developerInstructions=...); run/fork do
not re-send. Cloud executor has no such channel, so the worker
falls back to the v1 marked header in the in-memory dispatch
prompt only (DB prompt stays clean). Goal/coordinator tasks
still reject instructions.

Tests: test_agent_def_v2.py (7 tests) + updated v1 suite.

cloud/worker.py

@@ -134,6 +134,8 @@ import time
 import uuid
 from types import SimpleNamespace
 
+from cloud.agent_def import AgentDefinition as _AgentDefinition
+from cloud.agent_def import format_instructions_header as _agent_header
 from cloud.auth import AuthStore
 from cloud.coordinator import build_report as _coord_build_report
 from cloud.coordinator import DEFAULT_POLL_INTERVAL as _COORD_POLL_INTERVAL
@@ -412,6 +414,10 @@ def _task_from_row(row):
         "coordinator_parts": _coord_parse_parts(_col("coordinator_parts")),
         "parent_task_id": _col("parent_task_id"),
         "coordinator_part": _col("coordinator_part"),
+        # agent 定义 v2：老行缺列时 None = 未用定义，不炸。
+        "agent_name": _col("agent_name"),
+        "agent_version": _col("agent_version"),
+        "agent_instructions": _col("agent_instructions"),
     }
 
 
@@ -5864,6 +5870,21 @@ class Worker:
                                 task_id, decision.tier, decision.source,
                                 decision.applied))
             task["model"] = decision.chosen_model or self.model
+        # agent 指令通道（v2）：local/docker 由 sandbox 的
+        # --developer-instructions 派发，prompt 保持干净；通用
+        # cloud 执行器没有这个通道，只能在这里把 v1 的带标记
+        # 头拼进内存里的派发 prompt（只改本轮 task 副本，不写
+        # 库，库 prompt 保持干净），否则 cloud 任务会比 v1
+        # 倒退成指令全丢。已带头的 prompt 不重复拼。
+        if task.get("executor") == "cloud" \
+                and task.get("agent_instructions") \
+                and not (task.get("prompt") or "").startswith("[Agent:"):
+            task["prompt"] = _agent_header(
+                _AgentDefinition(
+                    name=task.get("agent_name") or "agent",
+                    version=task.get("agent_version"),
+                    instructions=task["agent_instructions"]),
+                task.get("prompt") or "")
         actual_model = task["model"]
         # 有限重试与真实 fallback 的优先级：取消不重试；一次尝试失败
         # 且可 fallback、且显式配了 fallback_models 时，优先真换模型