Module question 

Tool for asking structured questions to the user.

Question provides a way for LLM agents to ask the user for information in a structured, efficient way. Supports free-text and choice questions (single/multi select) and allows asking several questions at once to minimize back-and-forth.

Design principles

• Market-standard wire format: the schema mirrors Claude Code’s AskUserQuestion (and its ports — Qwen Code, OpenCode, Crush): a questions array of { question, header, options: [{label, description}], multiSelect } items. Models trained on any of those harnesses emit a valid call on the first attempt.
• Lenient decoding of model emission variants: plain-string options, string-encoded arrays, the multiple alias, and header-only items all parse — see types::QuestionItem’s custom deserializer. These are model behaviors, not alternate schemas.
• Batch questions: ask multiple questions in a single call.
• Purpose field: each question may explain why the information is needed.
• Recommendations by convention: the recommended option is the FIRST option carrying a " (Recommended)" label suffix — the convention models already know. The suffix is stripped wherever it appears and turned into the internal recommended badge.

Example

use cosh_tools::question::Question;
use cosh_tools::question::types::QuestionInput;

let q = Question::new();
let input: QuestionInput = serde_json::from_value(serde_json::json!({
    "questions": [{
        "question": "Which auth method?",
        "header": "Auth method",
        "options": [
            {"label": "OAuth 2.0 (Recommended)", "description": "Browser flow"},
            {"label": "API key", "description": "Static token"}
        ]
    }]
}))?;
let output = q.ask(&input)?;

Modules

types

Structs

Question

Tool for asking structured questions to the user.