boardx / boardx/workspacex

deep-agent:devapp 的 qwen3.8-max 默认开着 thinking,task 子代理每一步都慢

Open
#2,811 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
TypeScript
Stars
0
Forks
0
Avg merge
1h 7m
Merged PRs (30d)
969

Description

## 现象(devapp 实测,2026-09-06 人类反馈)
`task` 工具(同步子代理委托)处理时间很长;人类要求把 thinking 模式关掉。

## 根因
- devapp 实际部署的 deep-agent 模型是 `qwen3.8-max`(`.harness/state/deepagent-eval/2026-08-23-3d327c13/sse-and-thread-state-evidence-v2/01-sse-stream.txt` 里 140 条 `"model_name":"qwen3.8-max"`)。
- 关 thinking 的门控(#2504 / #2700)只对 `KERNEL_MODEL_THINKING_DISABLE_IDS` 集合里的模型发 `enable_thinking: false`,默认集合是 `qwen-plus,qwen3.7-plus`,不含 `qwen3.8-max`。
- 百炼文档:qwen3.5/3.6/3.7/3.8 系列是混合思考模型,`enable_thinking` **默认 true**(https://help.aliyun.com/zh/model-studio/deep-thinking)。于是主 agent 与所有子代理每一步都在思考。

## 处置
把 `qwen3.8-max` 补进 Python(`model.py`)/ TS(`configured-model-provider.ts`)两侧默认集合,并加 pytest 机械比对两侧一致。见关联 PR。

运维替代路径(不改代码):在 deploy.env 里设 `KERNEL_MODEL_THINKING_DISABLE_IDS=qwen-plus,qwen3.7-plus,qwen3.8-max`。

Contributor guide

No contributing guide indexed for this repository

Research direction

Start with model.py and configured-model-provider.ts, then inspect the existing KERNEL_MODEL_THINKING_DISABLE_IDS defaults and the related pytest coverage. Compare the two language-side collections and verify that qwen3.8-max is included consistently; done means the pytest mechanical comparison passes and the default disables thinking for this model.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, typescript
Domain
ai, backend
Issue type
Bug
Difficulty
3/5
Estimated time
1-2 days
Activity status
Active
Clarity
Clearly specified
Newbie friendliness
76/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.