Skip to content

Questions hang with no output #296

Closed

Description

@rmhubbert

Hey,

I'm not sure if this is related to the model (Qwen3 Coder Next), the Opencode API, or opencode.nvim, but I am seeing occassional issues with question asking.

About 50% of the time the model uses the question tool, I see the ? question prompt, but no questions are ever output. It just sits with Thinking.

I'm hosting my model locally, and don't see any errors in the vLLM logs. Also, I don't see the issue when using the Opencode TUI.

I'm running an up to date Arch Linux with the most recent versions of Neovim, Opencode, and opencode.nvim.

Output from checkhealth opencode

==============================================================================
opencode:                                                                   ✅

OpenCode CLI ~
- ✅ OK opencode command found
- ✅ OK opencode CLI version: 1.2.10 (>= 0.6.3 required)

OpenCode Server ~
- ✅ OK opencode server started successfully at http://127.0.0.1:38373
- ✅ OK opencode server is reachable
- ✅ OK opencode server configuration available
- ✅ OK opencode server shut down successfully

Dependencies ~
- ✅ OK plenary.nvim found

Configuration ~
- ✅ OK UI position: right
- ✅ OK Window width: 0.5
- ✅ OK Input min height: 0.1
- ✅ OK Input max height: 0.25
- ✅ OK Configuration loaded successfully

Environment ~
- ✅ OK Git repository detected
- ✅ OK Working directory: /home/hubby/code/go/csq

Optional Integrations ~
- ✅ OK telescope.nvim found (enhanced file picker available)
- mini.pick not found
- fzf-lua not found
- ✅ OK snacks.nvim found (enhanced file picker available)
- ✅ OK blink found (enhanced completion available)
- ✅ OK nvim-cmp found (enhanced completion available)

Finder CLI Tools ~
- ✅ OK Found file finder command: fd

Activity

  1. disrupted commented on Feb 24, 2026

    @disrupted
    Contributor

    does that happen in your regular session or only with child sessions (subagents)? If you could provide an event log that would be really helpful.

    enable both of these

    opencode.nvim/README.md

    Lines 309 to 310 in ee7b443

    enabled = false, -- Enable debug messages in the output window
    capture_streamed_events = false,

    then reproduce the issue and execute :lua require('opencode.ui.debug_helper').save_captured_events('opencode_missing_question.json')

    feel free to redact anything that you don't want to share in the event log

  2. rmhubbert commented on Feb 25, 2026

    @rmhubbert
    Author

    Thanks @disrupted, your help is very much appreciated.

    I've seen it in both regular and child sessions. Typically, since I enabled debugging, I've only seen it happen once. Unfortunately, that happened right at the end of a long session, so the exported debug logs were over 460MB is size, and cut off before the bug occurred. I'll keep debugging on, and hopefully next time it will catch it and be able to share it here!

    That did get me thinking, though, I do tend to see the issue more often at the end of a long running session. My typical workflow uses an orchestrator agent to delegate to research, planning, and task management subagents, and the final act for the orchestrator is to use the question tool to ask for next steps once the research report, implementation plan, and task list have been returned to it. I wonder if context size could be a factor here?

  3. rmhubbert commented on Feb 26, 2026

    @rmhubbert
    Author

    @disrupted Please find the log attached. Again, this happened right at the end of my agent workflow.

    opencode_missing_question.json.tar.gz

  4. khaitranhq commented on Feb 28, 2026

    @khaitranhq
    Contributor

    @rmhubbert Hope that #302 can resolve your issue

  5. rmhubbert commented on Feb 28, 2026

    @rmhubbert
    Author

    @rmhubbert Hope that #302 can resolve your issue

    Nice! Will update just now. Thanks for the heads up.

  6. disrupted commented on Mar 1, 2026

    @disrupted
    Contributor

    @rmhubbert I wasn't able to reproduce the issue with your event log. When I replay it I do see the final question "I have completed the research, ..." including the options to answer it.

    could it have been that you were inside a child session and then switched back to the main session? This is where I believe @khaitranhq's bugfix is right on!

  7. rmhubbert commented on Mar 1, 2026

    @rmhubbert
    Author

    @disrupted Aye, that's exactly what would have happened. The workflow in those logs has an orchestrator agent call a researcher, planner, and task manager subagent before the orchestrator itself asks the question of how to proceed. I tend to check the progress as it is happening, so I will have been switching the view between the orchestrator and sub agents during the session.

    I've updated, so will test today to see if my issue has been resolved by @khaitranhq's patch.

  8. rmhubbert commented on Mar 4, 2026

    @rmhubbert
    Author

    Just to note, I haven't seen a repeat of my issue since @khaitranhq's patch was added, so please feel free to close this one.

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

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions