Question tool error, ' stops/hangs execution completely in headless mode

[rwese]

Description

When I run opencode in a a loop it sometimes causes the Agent to ask questions,

question_tool_error.json

i use opencode run --agent ... , so questions are useless too, causing my loop/script to get stuck.

→ Asked 2 questions
Error: The question tool was called with invalid arguments: [
  {
    "origin": "string",
    "code": "too_big",
    "maximum": 30,
    "inclusive": true,
    "path": [
      "questions",
      0,
      "options",
      0,
      "label"
    ],
    "message": "Too big: expected string to have <=30 characters"
  },
  {
    "origin": "string",
    "code": "too_big",
    "maximum": 30,
    "inclusive": true,
    "path": [
      "questions",
      1,
      "options",
      0,
      "label"
    ],
    "message": "Too big: expected string to have <=30 characters"
  }
].
Please rewrite the input so it satisfies the expected schema

Plugins

No response

OpenCode version

OpenCode 1.1.31

Steps to reproduce

1. I was unable to force the error, i hope the exported session helps.

attempt, here it worked fine everytime:

we are debugging a opencode issue, it fails to propperly handle/ask very long choice question. 
write a 2 paragraph stories and ask which is better to verify this issue.
ask using the question too.

Screenshot and/or share link

https://opncd.ai/share/uF0512WZ

Operating System

MacOS Tahoe 26

Terminal

Ghostty, but was running in a shellscript loop

[github-actions]

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

• Question tools block execution in opencode run ... mode #9830: Question tools block execution in opencode run ... mode - directly related to headless mode blocking
• Error: The question tool was called with invalid arguments #8650: Error: The question tool was called with invalid arguments - similar validation error with question tool (too_big for label vs header)
• Question tool throws if header length is >= 12 chars #7616: Question tool throws if header length is >= 12 chars - related to character limit validation in question tool

Feel free to ignore if none of these address your specific case.

[rwese]

Expected behavior would be:

• question 'tool' error is returned to the model for rewriting the question
• during non-interactive terminal session, the question tool must not be available

[rwese]

Let me know if any of the options, or any other, you would like to see, I am happy to implement it, as it breaks my ralph. :(

[rwese]

Found the workaround, disable the tool, either globally or for a specific agent.

{
  "$schema": "https://opencode.ai/config.json",
  "permission": {
    "question": "deny"
  }
}

[rwese]

Happens also when using as ACP, in Zed for example:

I am unaware if this might be a implementation issue in the Client

Session Export:

acp_session.json

        {
          "id": "prt_beb3703de001lJ28nbZsvWaZ3o",
          "sessionID": "ses_414cd0464ffeKV5D33YzVhi6kL",
          "messageID": "msg_beb36fb5c0011c3DnH7gu0ENIp",
          "type": "tool",
          "callID": "call_function_qjwy0qfku09b_1",
          "tool": "question",
          "state": {
            "status": "running",
            "input": {
              "questions": [
                {
                  "header": "Skill check",
                  "multiple": false,
                  "options": [
                    {
                      "description": "A script or hook named skill-check or similar",
                      "label": "Custom script"
                    },
                    {
                      "description": "A lefthook hook that validates skill files",
                      "label": "Lefthook hook"
                    },
                    {
                      "description": "Something else not listed here",
                      "label": "Other"
                    }
                  ],
                  "question": "What specific \"skill check\" are you referring to? I don't see any skill-related pre-commit hooks in the current lefthook.yml configuration. Could you clarify where this check is defined?"
                }
              ]
            },
            "time": {
              "start": 1769177811182
            }
          }
        }

Bonus issue: The weird thing is, the agent did not have the 'question' permission, via cli they never ask.

[yajo]

Isn't this fixed in #11379?