Skip to content

使用subagent读取图片文件,模型返回非预期内容 #4876

@MachineXu

Description

@MachineXu

What happened?

创建一个subagent(比如图片分析专家),要求识别图片并描述图片内容;
主agent提交图片分析任务,调用图片分析subagent读取图片,subagent返回了和图片完全无关的内容;

已确认主agent使用相同模型读取图片,可以返回正确的图片描述;

What did you expect to happen?

希望subagent使用read_file读取图片,模型能够正确识别;

Client information

Client Information

Run qwen to enter the interactive CLI, then run the /about command.

$ qwen /about
  > /about

  ╭────────────────────────────────────────────────────────────────────────────────────────────────╮
  │                                                                                                │
  │ 状态                                                                                           │
  │                                                                                                │
  │ Qwen Code                        0.17.1 (09d0faf6b)                                            │
  │ 运行环境                         Node.js v24.14.0 / npm unknown                                │
  │ LSP                              disabled                                                      │
  │ 操作系统                         win32 x64 (10.0.19045)                                        │
  │                                                                                                │
  │ 认证                             API Key - openai                                              │
  │ Base URL                         http://192.168.199.229:3000/v1                                │
  │ 模型                             doubao-seed-2.0-pro                                           │
  │ 快速模型                         doubao-seed-2.0-pro                                           │
  │ 会话 ID                          e6bd9aad-1ffc-459f-98f1-838141bec040                          │
  │ 沙箱                             no sandbox                                                    │
  │ 代理                             no proxy                                                      │
  │ 内存使用                         131.5 MB                                                      │
  │                                                                                                │
  ╰────────────────────────────────────────────────────────────────────────────────────────────────╯

Login information

模型:doubao-seed-2.0-pro

Anything else we need to know?

openai要求图片内容必须在user context,subagent图片数据放在tool result,是否是这一点导致的的?

Metadata

Metadata

Assignees

Labels

category/coreCore engine and logiccategory/toolsTool integration and executionpriority/P2Medium - Moderately impactful, noticeable problemroadmap/subagents-toolsRoadmap: Subagents and tool systemstatus/need-informationMore information is needed to resolve this issue.type/bugSomething isn't working as expected

Type

No type
No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions