Repository navigation
LLM frequently exceeds question tool label character limit #10283
Description
Description
The question tool has a 30-character limit on option labels, but this constraint is not clearly communicated in the tool description. This causes LLMs to frequently exceed the limit on their first attempt, requiring a retry after seeing the validation error.
Current behavior
- Schema has
label: z.string().max(30) - Tool description says "add '(Recommended)' at the end of the label" for recommended options
- LLMs create labels like "Use existing patterns (Recommended)" (36 chars) → validation error
Example error:
Error: The question tool was called with invalid arguments: [
{
"code": "too_big",
"maximum": 30,
"path": ["questions", 0, "options", 0, "label"],
"message": "Too big: expected string to have <=30 characters"
}
]
Expected behavior
- Tool description should explicitly state the 30-character limit for labels and headers
- Recommendation indicator should be shorter to fit within limits (e.g.,
*instead of(Recommended))
Proposed fix
Update packages/opencode/src/tool/question.txt to:
- Add explicit character limits for labels (30 chars) and headers (30 chars)
- Change recommendation suffix from "(Recommended)" (13 chars) to " *" (2 chars)
Metadata
Metadata
Assignees
Labels
No labels
Activity
github-actions commentedon Jan 23, 2026
This issue might be a duplicate of existing issues. Please check:
Feel free to ignore if none of these address your specific case.
docs(tool): clarify question tool label character limits
Arasple commentedon Jan 24, 2026
same here, been running into this frequently
rmk40 commentedon Jan 26, 2026
PR resolves this and is a simple change, just needs to be approved.
git-malu commentedon Mar 8, 2026
currently situation: can't display the full text of the options
tredondo commentedon Mar 11, 2026
This issue is still present in v1.2.24, and important, as it severely cripples the LLM output for complex answers. Please prioritize.
aklajnert commentedon Mar 12, 2026
The proposed solution is wrong in my opinion. The TUI client is able to show a whole output of the command. Instead of limiting the model's response, just modify the GUI to either not truncate the text or allow to remove truncation on demand.
There is a workaround - if you open the context and expand the latest message, you can see the raw question with a whole content.