[BUG] AskUserQuestion tool call fires but cards don't render — v1.1.5749

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

When a skill invokes the AskUserQuestion tool, the tool call fires (visible in the UI as a collapsed entry labelled "Attempted to ask pre-generation questions for [task name] >") but no interactive radio-button cards are rendered. Claude detects no usable response and falls back to generating plain-text numbered questions instead.

What Should Happen?

The AskUserQuestion tool should render an interactive card UI with selectable radio-button options — one card per question — allowing the user to click their answer rather than type it. This is the behaviour described in Cowork's system prompt, which states the tool "renders proper radio-button UI cards."

Error Messages/Logs

No explicit error message. The only visible indicator is the collapsed tool call entry: "Attempted to ask pre-generation questions for [task name] >". No error is thrown — the tool call silently produces no rendered output, causing Claude to fall through to text fallback.

Steps to Reproduce

1. Open Cowork and start a new conversation
2. Trigger any installed skill that calls AskUserQuestion (e.g. a brand design skill)
3. Observe the response — a collapsed "Attempted to…" entry appears, followed by Claude generating plain-text questions instead of interactive cards

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

1.1.5749 (ecf3d9) — 2026-03-08

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Other

Additional Information

<img width="999" height="630" alt="Image" src="https://github.com/user-attachments/assets/f8183f70-e0ca-413c-a7b9-09719a3bd51e" />

AskUserQuestion is documented in Cowork's system prompt as an available tool but does not appear in Claude's callable function schema (no JSON schema definition exists for it, unlike TodoWrite which has a full schema). This means Claude cannot reliably detect the tool as available and defaults to text mode. The tool call does go out — confirmed by the "Attempted to…" UI label — but the Cowork client does not render a card response. Attaching screenshot showing the collapsed tool call entry and the resulting text fallback output.