[BUG] SendMessage routing failure causes synthetic heartbeat loop burning ~5400 wasted tool calls over 7 hours

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?

A dispatched subagent (a plan action for a Jira ticket) did real, substantive work for only ~23 minutes — grounding against code, authoring a plan, dispatching a 3-reviewer panel including one that found and got a revision for a real blocking finding. At the point it sent the revised plan back to one reviewer sub-agent for round-2 re-review via SendMessage, that reply never arrived through the expected channel. Instead of timing out or surfacing the stall, the agent began manufacturing a continuous synthetic "not idle" heartbeat — echo idle / sleep N; echo ok / echo "pending waits armed" — firing every 5-8 seconds for the remaining ~7 hours until manually killed. Of 5,486 total tool calls in the transcript, 5,431 (99%) were this content-free heartbeat loop. Each call, despite trivial content, is a full tool-call round-trip that resends the full context, so cost scaled with call count rather than actual work done.

What Should Happen?

Claude should stop retrying vs getting stuck in an expensive retry loop.

Error Messages/Logs

Steps to Reproduce

Dispatch an agent that itself dispatches a nested sub-agent and awaits its reply via SendMessage, where the sub-agent's reply-routing back to the parent fails. The parent does not time out; it enters an apparently indefinite synthetic-heartbeat loop instead of surfacing the failure or exiting.

Claude Model

None

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.252

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

IntelliJ IDEA terminal

Additional Information

_No response_