fix: session resume, config_set schema, metrics chart, subagent logging

- Fix session resume: remove messages.length>1 guard so sessionStorage
  ID is used as resumeSessionId on first message (prevents new session
  creation when user sends "继续")
- Fix config_set tool schema: add missing `items: {}` to array type in
  oneOf (Azure OpenAI strict validation rejects it otherwise)
- Metrics chart: use actual target set filename from trigger text
  instead of generic "target_subset" key
- Add sub-agent skill call logging to agent_runtime.py (full
  prompt/response, no truncation)
- Various metric enrichment and dedup fixes in server.py

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/agent_runtime.py

@@ -154,6 +154,119 @@ def _prepend_runtime_context(prompt: str, runtime_context: str | None) -> str:
     )
 
 
+def _log_child_skill_calls(
+    child_result: 'AgentRunResult',
+    child_agent: 'LocalCodingAgent',
+    subtask_label: str,
+) -> None:
+    """Extract Skill tool calls from a child agent's transcript and write to a log file.
+
+    Logs each Skill invocation with the args sent and the agent's response,
+    so we can audit whether sub-agents actually called skills like label-master.
+    """
+    from .agent_types import AgentRunResult  # noqa: F811
+
+    transcript = child_result.transcript
+    if not transcript:
+        return
+    log_dir = child_agent.runtime_config.session_directory / '_subagent_logs'
+    try:
+        log_dir.mkdir(parents=True, exist_ok=True)
+    except OSError:
+        return
+    session_id = child_result.session_id or 'unknown'
+    log_path = log_dir / f'{session_id}.jsonl'
+
+    skill_calls: list[dict[str, object]] = []
+    # Build a map of tool_call_id -> skill args from assistant messages with tool_calls
+    pending_skills: dict[str, dict[str, str]] = {}
+    for entry in transcript:
+        if not isinstance(entry, dict):
+            continue
+        role = entry.get('role')
+        if role == 'assistant':
+            tool_calls = entry.get('tool_calls')
+            if isinstance(tool_calls, list):
+                for tc in tool_calls:
+                    if not isinstance(tc, dict):
+                        continue
+                    fn = tc.get('function') or {}
+                    if not isinstance(fn, dict):
+                        continue
+                    if fn.get('name') == 'Skill':
+                        tc_id = tc.get('id', '')
+                        raw_args = fn.get('arguments', '{}')
+                        try:
+                            args_dict = json.loads(raw_args) if isinstance(raw_args, str) else raw_args
+                        except (json.JSONDecodeError, TypeError):
+                            args_dict = {'raw': raw_args}
+                        pending_skills[tc_id] = {
+                            'skill': args_dict.get('skill', ''),
+                            'args': args_dict.get('args', ''),
+                        }
+        elif role == 'tool':
+            tc_id = entry.get('tool_call_id', '')
+            if tc_id in pending_skills:
+                skill_info = pending_skills.pop(tc_id)
+                skill_calls.append({
+                    'skill': skill_info['skill'],
+                    'args': skill_info['args'],
+                    'prompt_injected': entry.get('content') or '',
+                    'ts': datetime.now(timezone.utc).isoformat(),
+                })
+
+    # Also capture the assistant responses that follow skill injections
+    # by looking at assistant messages after tool messages
+    assistant_after_skill: list[str] = []
+    saw_skill_tool = False
+    for entry in transcript:
+        if not isinstance(entry, dict):
+            continue
+        role = entry.get('role')
+        if role == 'tool' and entry.get('tool_call_id', '') in [
+            tc.get('id', '') for tc in _all_skill_tool_call_ids(transcript)
+        ]:
+            saw_skill_tool = True
+        elif role == 'assistant' and saw_skill_tool:
+            content = entry.get('content', '')
+            if content:
+                assistant_after_skill.append(content)
+            saw_skill_tool = False
+
+    # Merge assistant responses into skill_calls
+    for i, response_text in enumerate(assistant_after_skill):
+        if i < len(skill_calls):
+            skill_calls[i]['agent_response'] = response_text
+
+    if not skill_calls:
+        return
+    try:
+        with open(log_path, 'a', encoding='utf-8') as f:
+            for call in skill_calls:
+                call['subtask_label'] = subtask_label
+                f.write(json.dumps(call, ensure_ascii=False) + '\n')
+    except OSError:
+        pass
+
+
+def _all_skill_tool_call_ids(transcript: tuple[dict[str, object], ...]) -> list[dict[str, str]]:
+    """Collect all tool_call entries for Skill from assistant messages."""
+    results = []
+    for entry in transcript:
+        if not isinstance(entry, dict) or entry.get('role') != 'assistant':
+            continue
+        tool_calls = entry.get('tool_calls')
+        if not isinstance(tool_calls, list):
+            continue
+        for tc in tool_calls:
+            if not isinstance(tc, dict):
+                continue
+            fn = tc.get('function') or {}
+            if isinstance(fn, dict) and fn.get('name') == 'Skill':
+                results.append(tc)
+    return results
+
+
 @dataclass
 class LocalCodingAgent:
     model_config: ModelConfig
@@ -3052,6 +3165,11 @@ class LocalCodingAgent:
                     managed_child_index=index,
                     managed_label=subtask_label,
                 )
+                if self.tool_context.jupyter_runtime is not None:
+                    child_agent.tool_context = replace(
+                        child_agent.tool_context,
+                        jupyter_runtime=self.tool_context.jupyter_runtime,
+                    )
                 if group_id is not None and child_agent.managed_agent_id is not None:
                     self.agent_manager.register_group_child(
                         group_id,
@@ -3118,9 +3236,11 @@ class LocalCodingAgent:
                             break
                         continue
                     child_result = child_agent.resume(child_prompt, stored_child_session)
+                    _log_child_skill_calls(child_result, child_agent, subtask_label)
                     resume_used = True
                 else:
                     child_result = child_agent.run(child_prompt)
+                _log_child_skill_calls(child_result, child_agent, subtask_label)
                 if group_id is not None and child_agent.managed_agent_id is not None:
                     self.agent_manager.register_group_child(
                         group_id,

src/agent_tools.py

@@ -340,7 +340,7 @@ def default_tool_registry() -> dict[str, AgentTool]:
                             {'type': 'number'},
                             {'type': 'integer'},
                             {'type': 'boolean'},
-                            {'type': 'array'},
+                            {'type': 'array', 'items': {}},
                             {'type': 'object'},
                             {'type': 'null'},
                         ]

src/jupyter_runtime.py

@@ -268,10 +268,12 @@ class JupyterRuntimeSession:
                 '远端 jupyter pod 未挂载 /mnt/wangsenhao 或不可写：'
                 + (probe.stdout.strip() or probe.stderr.strip() or 'unknown error')
             )
+        ws_output = f'{shlex.quote(self.binding.workspace_cwd)}/output'
+        chat_output_link = f'{shlex.quote(chat_root)}/output'
         result = self.run_command(
             (
                 'mkdir -p '
-                f'{shlex.quote(self.binding.workspace_cwd)}/output '
+                f'{ws_output} '
                 f'{shlex.quote(self.binding.workspace_cwd)}/input '
                 f'{shlex.quote(self.binding.workspace_cwd)}/scratchpad '
                 f'{shlex.quote(self.binding.workspace_root)}/.runtime/uploads '
@@ -279,7 +281,12 @@ class JupyterRuntimeSession:
                 f'{shlex.quote(self.account_runtime_root)}/python '
                 f'{shlex.quote(chat_root)}/results '
                 f'{shlex.quote(chat_root)}/sft_output '
-                f'{shlex.quote(chat_root)}/scripts'
+                f'{shlex.quote(chat_root)}/scripts && '
+                f'if [ -d {chat_output_link} ] && [ ! -L {chat_output_link} ]; then '
+                f'cp -a {chat_output_link}/* {ws_output}/ 2>/dev/null; '
+                f'rm -rf {chat_output_link}; '
+                f'fi && '
+                f'ln -sfnT {ws_output} {chat_output_link}'
             ),
             timeout_seconds=timeout_seconds,
             max_output_chars=4000,