AskUserQuestion UI not rendered when called after Skill tool invocation

[ArtemYurov]

Bug Description

AskUserQuestion tool sometimes fails to render the question UI. Instead of showing the question with options, it immediately returns "User answered Claude's questions:" with an empty answer.

Steps to Reproduce

1. Create a skill (slash command) that internally calls AskUserQuestion
2. Invoke the skill via /skill-name with arguments
3. The skill processes arguments and calls AskUserQuestion with options
4. Expected: UI displays the question with selectable options
5. Actual: No question UI shown, empty answer returned immediately

Example Call

{
  "questions": [{
    "question": "To be or not to be?",
    "header": "Hamlet",
    "options": [
      {"label": "To be", "description": "Choose existence"},
      {"label": "Not to be", "description": "Choose non-existence"}
    ],
    "multiSelect": false
  }]
}

Screenshot

The UI shows:

● User answered Claude's questions:
  └

No question text, no options displayed. Answer is empty.

Additional Context

• The bug appears to occur specifically when AskUserQuestion is called shortly after a Skill tool invocation in the same conversation turn
• A subsequent AskUserQuestion call (without Skill context) in the next turn rendered correctly
• This may be related to the timing/sequencing of tool calls after skill expansion
• Platform: macOS (Darwin 24.6.0)
• Claude Code CLI

[github-actions]

Found 3 possible duplicate issues:

1. AskUserQuestion silently returns empty answers when called inside plugin skills #29547
2. AskUserQuestion auto-submits without waiting for user input in skill/command context #29674
3. AskUserQuestion auto-submits with empty response when called from custom commands/skills #29690

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

[bopm]

After pushing around /debug skill to figure this out, here is what it given me:

⏺ Root cause found: tengu_streaming_tool_execution2: true

  This experimental feature flag enables tools to execute while the response is still streaming, rather than waiting for the full response to complete first. It's a performance optimization for tools like Read, Write, Bash — they can run
  concurrently with streaming.

  But AskUserQuestion is a blocking UI tool. It needs to:
  1. Pause everything
  2. Render a dialog
  3. Wait for user input
  4. Return the result

  With tengu_streaming_tool_execution2 active, AskUserQuestion fires during streaming — the UI context isn't in a state to show a blocking dialog. The tool executes, gets no input (or auto-returns empty), and the stream continues
  immediately. That's why only 22ms passed and why PostToolUse never fired — the dialog lifecycle never actually ran.

  This is a bug: tengu_streaming_tool_execution2 is incompatible with AskUserQuestion.

  To confirm/fix: Disable the flag in .claude.json:
  "tengu_streaming_tool_execution2": false

  Then retry the skill — AskUserQuestion should block properly and show the dialog.

[bopm]

And additional information:

it's clean room environment, who enabled that?

⏺ Searched for 2 patterns (ctrl+o to expand)

⏺ Bash(python3 -c "
      import json…)
  ⎿  cachedGrowthBookFeatures: dict

⏺ cachedGrowthBookFeatures — that's the source.

  GrowthBook is the A/B testing / feature flag platform Anthropic uses for Claude Code. The tengu_streaming_tool_execution2: true flag was remotely enabled by Anthropic and cached locally in .claude.json. You didn't set it — it was pushed
  to your account/install as part of a staged rollout or experiment.

  So the full chain:
  1. Anthropic's GrowthBook service pushed tengu_streaming_tool_execution2: true to your Claude Code install
  2. Claude Code cached it in .claude.json under cachedGrowthBookFeatures
  3. The flag caused tools to execute during streaming
  4. AskUserQuestion — which needs to block and show a dialog — fired during streaming and returned empty immediately
  5. Skill failed silently

  This is a bug Anthropic needs to fix: tengu_streaming_tool_execution2 should either exclude AskUserQuestion from streaming execution, or AskUserQuestion should detect this context and handle it correctly.

  You can report it at https://github.com/anthropics/claude-code/issues with the flag name and this behavior.

[cristianuibar]

use #29674 as the most relevant one. close this issue.

[bopm]

@cristianuibar my debug results are way more detailed than any of present in that issue of yours. Thoughts?