子agent调用报错

### Problem summary

Task Succeeded. Result: LLM request failed: Error code: 400 - {'error': {'message': 'System message must be at the beginning.', 'type': 'Ba

### Expected behavior

子agent执行成功

### Actual behavior

我在调试后端代码的时候，我留意到子agent的message拼接顺序有问题，具体报错如下：
================================== Ai Message ==================================
...

现在开始工作...
Tool Calls:
  read_file (chatcmpl-tool-b4ae020cd7799b40)
 Call ID: chatcmpl-tool-b4ae020cd7799b40
  Args:
    description: Load writing style guidelines
    path: /mnt/skills/custom/patent-navigation-report/writing_style/guidelines.md
  read_file (chatcmpl-tool-8e84c3daecada076)
 Call ID: chatcmpl-tool-8e84c3daecada076
  Args:
    description: Load chart specification
    path: /mnt/skills/custom/patent-navigation-report/plotting/chart_specification.md
  read_file (chatcmpl-tool-908f835968f3c2cb)
 Call ID: chatcmpl-tool-908f835968f3c2cb
  Args:
    description: Load general methodology
    path: /mnt/skills/custom/patent-navigation-report/methodology/general_methodology.md
  task (chatcmpl-tool-9586c236eb5954cd)
 Call ID: chatcmpl-tool-9586c236eb5954cd
  Args:
    description: Research global CF textile patents
    prompt: Research the latest information on carbon fiber textile patents globally and in China. I need:


...
Use web_search and web_fetch tools to gather this information. Be thorough and cite your sources.
    subagent_type: general-purpose
  task (chatcmpl-tool-b581e9d9dc23a3bb)
 Call ID: chatcmpl-tool-b581e9d9dc23a3bb
  Args:
    description: Research CF textile companies
    prompt: Research the major companies in the carbon fiber textile industry, focusing on their patent portfolios and competitive positions. I need detailed information on:

....

Search the web for recent news, company reports, and patent database information. Provide specific data and cite sources.
    subagent_type: general-purpose
  task (chatcmpl-tool-8191dc82aa4faa8a)
 Call ID: chatcmpl-tool-8191dc82aa4faa8a
  Args:
    description: Research CF textile market policies
    prompt: Research the market size, industry policies, and technology trends for carbon fiber textiles globally and in China. I need:

...

Search the web for recent industry reports, government policy documents, and market research. Provide specific numbers and cite sources.
    subagent_type: general-purpose
/data2/deer-flow/backend/.venv/lib/python3.12/site-packages/pydantic/functional_validators.py:835: UserWarning: Pydantic serializer warnings:
  PydanticSerializationUnexpectedValue(Expected `none` - serialized value may not be as expected [field_name='context', input_value={'thread_id': 'debug-thre...sandbox_id': '761c3b10'}, input_type=dict])
  function=lambda v, h: h(v), schema=original_schema
/data2/deer-flow/backend/.venv/lib/python3.12/site-packages/pydantic/main.py:475: UserWarning: Pydantic serializer warnings:
  PydanticSerializationUnexpectedValue(Expected `none` - serialized value may not be as expected [field_name='context', input_value={'thread_id': 'debug-thre...sandbox_id': '761c3b10'}, input_type=dict])
  return self.__pydantic_serializer__.to_python(
================================= Tool Message =================================
Name: task

Task Succeeded. Result: LLM request failed: Error code: 400 - {'error': {'message': 'System message must be at the beginning.', 'type': 'BadRequestError', 'param': None, 'code': 400}}




### Operating system

Linux

### Platform details

_No response_

### Python version

_No response_

### Node.js version

_No response_

### pnpm version

pnpm 10.33.1

### uv version

_No response_

### How are you running DeerFlow?

Local (make dev)

### Reproduction steps

make dev

### Relevant logs

```shell
我查看了后端日志如下：
2026-05-02 14:55:23 - deerflow.subagents.executor - INFO - [trace=4b2574c4] Subagent general-purpose captured AI message #1
2026-05-02 14:55:23 - deerflow.subagents.executor - INFO - [trace=4b2574c4] Subagent general-purpose completed async execution
2026-05-02 14:55:23 - deerflow.subagents.executor - INFO - [trace=4b2574c4] Subagent general-purpose final messages count: 24
2026-05-02 14:55:23 - httpx - INFO - HTTP Request: POST http://127.0.0.1:23333/v1/chat/completions "HTTP/1.1 400 Bad Request"
2026-05-02 14:55:23 - deerflow.agents.middlewares.llm_error_handling_middleware - WARNING - LLM call failed after 1 attempt(s): Error code: 400 - {'error': {'message': 'System message must be at the beginning.', 'type': 'BadRequestError', 'param': None, 'code': 400}}
Traceback (most recent call last):
  File "/data2/deer-flow/backend/packages/harness/deerflow/agents/middlewares/llm_error_handling_middleware.py", line 266, in awrap_model_call
    response = await handler(request)
               ^^^^^^^^^^^^^^^^^^^^^^
  File "/data2/deer-flow/backend/.venv/lib/python3.12/site-packages/langchain/agents/factory.py", line 1330, in _execute_model_async
    output = await model_.ainvoke(messages)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/data2/deer-flow/backend/.venv/lib/python3.12/site-packages/langchain_core/runnables/base.py", line 5752, in ainvoke
    return await self.bound.ainvoke(
           ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/data2/deer-flow/backend/.venv/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 494, in ainvoke
    llm_result = await self.agenerate_prompt(
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/data2/deer-flow/backend/.venv/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 1760, in agenerate_prompt
    return await self.agenerate(
           ^^^^^^^^^^^^^^^^^^^^^
  File "/data2/deer-flow/backend/.venv/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 1718, in agenerate
    raise exceptions[0]
  File "/data2/deer-flow/backend/.venv/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 2053, in _agenerate_with_cache
    result = await self._agenerate(
             ^^^^^^^^^^^^^^^^^^^^^^
  File "/data2/deer-flow/backend/.venv/lib/python3.12/site-packages/langchain_openai/chat_models/base.py", line 1923, in _agenerate
    _handle_openai_bad_request(e)
  File "/data2/deer-flow/backend/.venv/lib/python3.12/site-packages/langchain_openai/chat_models/base.py", line 1918, in _agenerate
    raw_response = await self.async_client.with_raw_response.create(
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/data2/deer-flow/backend/.venv/lib/python3.12/site-packages/openai/_legacy_response.py", line 384, in wrapped
    return cast(LegacyAPIResponse[R], await func(*args, **kwargs))
                                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/data2/deer-flow/backend/.venv/lib/python3.12/site-packages/openai/resources/chat/completions/completions.py", line 2714, in create
    return await self._post(
           ^^^^^^^^^^^^^^^^^
  File "/data2/deer-flow/backend/.venv/lib/python3.12/site-packages/openai/_base_client.py", line 1913, in post
    return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/data2/deer-flow/backend/.venv/lib/python3.12/site-packages/openai/_base_client.py", line 1698, in request
    raise self._make_status_error_from_response(err.response) from None
openai.BadRequestError: Error code: 400 - {'error': {'message': 'System message must be at the beginning.', 'type': 'BadRequestError', 'param': None, 'code': 400}}
```

### Git state

commit eba3b9e18d797dfc42b9cc8610781fa941755731 (HEAD -> main, origin/main, origin/HEAD)
Author: He Wang <wanghechn@qq.com>
Date:   Thu Apr 30 22:27:14 2026 +0800

    fix(config): unify log_level from config.yaml across Gateway and debug entry points (#2601)
    
    Centralize log level parsing in `logging_level_from_config()` and
    application in `apply_logging_level()` within `deerflow.config.app_config`.
    
    - Gateway lifespan applies configured log level on startup
    - `debug.py` uses shared helpers instead of local duplicates
    - `apply_logging_level()` targets only `deerflow`/`app` logger hierarchies
      so third-party library verbosity is not affected; root handler levels
      are only lowered (never raised) to allow configured loggers through
      without suppressing third-party output; root logger level is not modified
    - Config field description updated to clarify scope
    - Tests save/restore global logging state to avoid test pollution
    
    Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>

### Additional context

_No response_

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

子agent调用报错 #2693

Problem summary

Expected behavior

Actual behavior

Operating system

Platform details

Python version

Node.js version

pnpm version

uv version

How are you running DeerFlow?

Reproduction steps

Relevant logs

Git state

Additional context

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

子agent调用报错 #2693

Description

Problem summary

Expected behavior

Actual behavior

Operating system

Platform details

Python version

Node.js version

pnpm version

uv version

How are you running DeerFlow?

Reproduction steps

Relevant logs

Git state

Additional context

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions