[BUG] Text-only status messages mid-task silently end the turn and hand control back to the user
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 Claude emits a text-only message between tool calls (e.g., a short "let me check X" narration), and no tool call follows in the same turn, the turn terminates. Control is returned to the user as if the task were complete, even though Claude intended only a pause. The user must then send a follow-up message ("continue", "why did you stop?") just to resume work that was already in progress.
What Should Happen?
Narrative text between tool calls should not, on its own, end a turn. Either the turn should remain open until
Claude explicitly signals completion, or Claude should be trained/instructed to never emit mid-task narration
without chaining it to a tool call in the same message. From the user's perspective, pausing for thought should not
look identical to finishing the task.
Error Messages/Logs
Search(_____)
⎿ Found 5 files
✻ Churned for 38s
> why did you stop?
● I didn't stop, just pausing to map all call sites before editing. Continuing now.
Steps to Reproduce
- Give Claude a multi-step task (e.g., rename a record field across several files after verifying usages).
- Observe Claude running a Grep or similar discovery tool.
- Claude responds with a short text update (e.g., "Found the usages, now editing the files.") without a tool call
attached.
- The turn ends. The user is forced to type something to resume.
- Repeat: even after resuming, the same pattern can recur on the next discovery step UNLESS explicitly told to chain tool use.
Claude Model
Opus
Is this a regression?
Yes, this worked in a previous version
Last Working Version
_No response_
Claude Code Version
2.1.114
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
Windows Terminal
Additional Information
- Workaround added to project CLAUDE.md: "Never end a turn with a text-only message mid-task. Chain narration with the next tool call. Only end a turn when the task is complete or genuinely blocked."
- Happened repeatedly within a single session.
This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗