Agent asks user a question but proceeds with actions in the same response without waiting for answer
Bug
When Claude Code decides to ask the user a clarifying question (e.g., "Want me to go ahead?"), it sometimes includes tool calls (edits, writes) in the same response, executing the action without waiting for the user's answer. The question becomes rhetorical even though it was intended as a gate.
Root cause (self-reported)
Text and tool calls are generated as a single response unit. Writing a question doesn't create an actual decision point to suppress tool calls in the same message. The model needs to commit to either asking (response with only text) or acting (response with text + tool calls), not both.
Repro
Ask Claude Code to make a non-trivial change where it's uncertain enough to ask for confirmation. It may output a question and then immediately execute the change in the same response.
Expected behavior
If the response contains a question directed at the user, there should be no tool calls in that same response. The agent should wait for the user's reply before taking action.
This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗