[MODEL] Assistant scheduled a wakeup via ScheduleWakeup purely to poll a background Agent-tool subagent, contradicting the tool's explicit guidance
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
Other unexpected behavior
What You Asked Claude to Do
Description:
While waiting on a background subagent (launched via the Agent tool) to finish an investigation, the assistant called ScheduleWakeup with a short delay and the <<autonomous-loop-dynamic>> sentinel solely to be notified when the subagent completed. This is explicitly called out as incorrect in ScheduleWakeup's own tool description: "Do NOT schedule a short-interval wakeup to poll for background work you started — when harness-tracked work finishes, you are re-invoked automatically." Since no /loop was actually active in the session, this spun up an unintended autonomous-loop tick cycle, which then had to be explicitly stopped (stop: true) on the very next tick once the model noticed the mistake.
What Claude Actually Did
Actual: The assistant called it anyway, then had to self-correct one tick later.
Note: This is a model-behavior issue, not necessarily a tool/product defect — flagging in case it's a useful data point for instruction-following evaluation.
Expected Behavior
Expected: The assistant should never invoke ScheduleWakeup purely to poll a background task it already knows will notify it automatically on completion — this should be caught by the model reading (and following) the tool's own description before calling it.
Files Affected
Permission Mode
Accept Edits was OFF (manual approval required)
Can You Reproduce This?
Haven't tried to reproduce
Steps to Reproduce
- Have the assistant launch a background subagent via the Agent tool.
- Ask it to wait for the result.
- Observe whether it calls ScheduleWakeup (with the loop sentinel) merely to be notified of completion, rather than simply waiting for the automatic <task-notification>.
Claude Model
Sonnet
Relevant Conversation
Impact
Low - Minor inconvenience
Claude Code Version
2.1.267 (Claude Code)
Platform
Anthropic API
Additional Context
_No response_