Skip to content

当输出出现tool call的时候,对话被终止 #8877

Closed as not planned
Closed as not planned

Description

@zt449569708

Description

在模型对话输出的时候,如”Thinking: 您说得对!我需要重新检查。让我删除之前的所有修改文件,重新解压原始文档,然后仔细诊断问题:<tool_call>bash<arg_key>command</arg_key><arg_value> ...“,当出现“<tool_call>bash<arg_key>command....”的时候对话中断,不再继续输出。今天更新到1.1.23才出现的,之前使用都没出现。

Plugins

Oh My OpenCode

OpenCode version

1.1.23

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response

Activity

  1. github-actions commented on Jan 16, 2026

    @github-actions
    Contributor

    This issue might be a duplicate of existing issues. Please check:

    Both issues describe conversations being interrupted when tool calls appear in the output, similar to your report. Feel free to ignore if your specific case differs.

  2. bi-kai commented on Jan 23, 2026

    @bi-kai

    same issue, with version 1.1.27

    mark & focus

  3. imjoey commented on Feb 5, 2026

    @imjoey

    same issue, version 1.1.51 with GLM-4.7 model

  4. kade06 commented on Feb 28, 2026

    @kade06

    same issue, version 1.2.15 with GLM-5 model。 end with </arg_value></tool_call>

  5. liuyibo123 commented on Mar 2, 2026

    @liuyibo123

    same issue ,version 1.2.15 with GLM-5 。</arg_value></tool_call> shows.

  6. SupetZYK commented on Mar 7, 2026

    @SupetZYK

    same issue version 1.2.20 with GLM-5 </arg_value></tool_call> shows.

  7. wangeek commented on Mar 17, 2026

    @wangeek

    same issue version 1.2.26 with GLM-4.7 </arg_value></tool_call> shows.

  8. MNCloudwerksTechnology commented on Mar 17, 2026

    @MNCloudwerksTechnology

    Same issue version v1.2.27 and qwen3.5-35B-A3B

    Stops directly after thinking block or after toolcalling. Happens in longer tasks more often. (+100k Context)

    `
    ┃ Thinking: The test failure "should emit logo click event for dashboard FAILED" suggests an issue with how logoClick is emitting. Let me check the platform component's test to see if that's the failing test. I'll skip it for now and run all platform tests: src/app/pages/platform/platfo +18 -42
    ┃
    ┃ <tool_call>
    ┃ <function=bash>
    ┃ <parameter=command>
    ┃ npm test -- --browsers=ChromeHeadless --watch=false --include=src/app/pages/platform/platform.component.spec.ts 2>&1 | grep -E "(FAILED|SUCCESS|TOTAL|passed)" | tail -10
    ┃
    ┃ <parameter=description>
    ┃ Run platform tests
    ┃
    ┃ <parameter=timeout>
    ┃ 120000
    ┃
    ┃
    ┃ </tool_call>

     ▣  Build · Qwen3.5-35B-A3B · 18.7s  
    

    `

  9. hashelq commented on Mar 23, 2026

    @hashelq

    If you use qwen (or a fork of it, like omnicoder) -> ollama/ollama#14492

    Interestingly, seems like switching to
    
    RENDERER qwen3-coder
    PARSER qwen3-coder
    
    for qwen3.5:9b fixes the tool calling problems for a trade off of broken think tag parsing.
    
  10. removed
    bugSomething isn't working
    on May 3, 2026
  11. github-actions commented on Jul 6, 2026

    @github-actions
    Contributor

    To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions