[Sonnet 4.6] AskUserQuestion returns without capturing user input during skill execution (intermittent)

Preflight Checklist

• [x] I have searched existing issues for similar behavior reports
• [x] This report does NOT contain sensitive information (API keys, passwords, etc.)

Type of Behavior Issue

Claude ignored my instructions or configuration

What You Asked Claude to Do

When a skill calls the AskUserQuestion tool, it occasionally returns immediately without waiting for or capturing the user's response. The skill then either continues with no answer recorded or falls back to text-based prompting.

This appears to be intermittent — it does not happen on every call. In one observed session, the first AskUserQuestion call in a skill returned empty, while a second call in the same skill execution worked correctly.

What Claude Actually Did

• AskUserQuestion renders (or appears to render) the question
• The tool returns immediately with no answer captured (empty result)
• The skill proceeds without user input

Expected Behavior

• AskUserQuestion blocks until the user selects an option or provides input
• The selected value is returned and available to the skill for subsequent steps

Files Affected

NA

Permission Mode

Accept Edits was OFF (manual approval required)

Can You Reproduce This?

Sometimes (intermittent)

Steps to Reproduce

1. Create or use a skill that includes AskUserQuestion in allowed-tools
2. Invoke the skill
3. When the skill calls AskUserQuestion, observe whether the UI captures the response

A minimal skill to reproduce:

---
name: test-ask
description: "Test AskUserQuestion behavior"
allowed-tools: ["AskUserQuestion"]
---

Ask the user using AskUserQuestion:
> What would you like to do?

Options: Option A, Option B

Claude Model

Sonnet

Relevant Conversation

Confidential

Impact

Medium - Extra work to undo changes

Claude Code Version

2.1.63

Platform

Anthropic API

Additional Context

What Was Investigated / Ruled Out

• Skill definition correctness: AskUserQuestion is properly listed in allowed-tools and called explicitly in skill instructions — not a skill authoring issue
• settings.json flags: No non-interactive, autoApprove, or equivalent settings present
• Intentional skip logic: The skill had no conditional logic that would bypass the AskUserQuestion call in this code path
• Model behavior (partial): The model does not appear to be "skipping" the call — the tool is called, but returns without data
• Reproducibility: Intermittent. Second AskUserQuestion call in the same skill execution worked correctly

---

Environment

• Claude Code version: 2.1.63
• Model: claude-sonnet-4-6 (issue not observed with claude-opus-4-6)
• OS: macOS
• effortLevel: "medium" (explicitly set in settings.json)
• Skill location: Local user skill (~/.claude/skills/)

---

More information

The issue was first noticed around 2026-03-02, specifically when using claude-sonnet-4-6. It was not observed with claude-opus-4-6, suggesting this may be model-specific behavior.

The issue was observed specifically on the first AskUserQuestion call when a skill is initially invoked — suggesting a possible initialization/timing issue within the skill execution context. Subsequent calls
in the same skill run correctly.

---
_[PS: This bug report was drafted with the help of Claude Code but has been reviewed and approved by the user.]_