Agent View: undocumented behavior — are run_in_background child processes killed when the supervisor reaps an idle session after ~1h?

Summary

The Agent View docs state that once a session finishes and sits unattached for ~1 hour, the supervisor stops its process to free resources, then restarts a fresh process on next attach/peek/reply. What happens to background child processes spawned via run_in_background during that reaping window is not documented.

Scenario

1. A background session's agent launches a long-running task (e.g. a multi-hour simulation) via run_in_background: true.
2. The agent's turn ends, so the session is "finished"/idle while the detached child keeps running.
3. The session sits unattached for ~1 hour and the supervisor stops the parent Claude Code process.

Questions / gap

• Does the supervisor kill the background child process when it stops the parent? (Normally a child would die with its parent.)
• Does an active tracked background task prevent the ~1h reaping? The UI already shows Cannot open agents — N background task(s) running, implying background tasks are tracked.
• If the child is killed, the harness's "re-invoke the agent when a background task completes" contract can never fire — the task and its completion callback are silently lost.

Requests

1. Document the exact interaction between the ~1h supervisor reaping and in-flight run_in_background tasks.
2. Ideally handle it: keep a session exempt from reaping while it has tracked background tasks still running, so long jobs and the completion callback survive.

Affected docs

• https://code.claude.com/docs/en/agent-view
• https://code.claude.com/docs/en/tools-reference

Filed via Claude Code.