[Bug] Background bash tasks killed after ~30 minutes during active sessions with context compaction

Bug Description
Background Bash tasks (run_in_background: true) are silently killed roughly every 30 minutes in a long-running session, while persistent Monitor tasks in the same session survive for days. Details: - Session has been running/resumed over ~6 days with heavy activity and frequent context compaction. - Every backgrounded shell (a supervisor script for a long-lived listener process) gets status "killed" with a "was stopped" task notification ~30-40 min after start, during active periods. During idle overnight periods the same shell survived 4+ hours, so the kills appear correlated with session activity (possibly compaction), not wall-clock time. - No settings configured that would explain this: no BASH_DEFAULT_TIMEOUT_MS / BASH_MAX_TIMEOUT_MS / background-related keys in user or project settings.json. - tools-reference.md and env-vars.md do not document any lifetime for background tasks, nor their interaction with context compaction, --resume, or auto-update. - Impact: any long-lived supervisor pattern built on run_in_background requires manual re-arming dozens of times per day (38+ times in two days in our case). Requests: 1. Document the background-task lifecycle (when/why they are terminated). 2. Provide a setting to opt long-lived tasks out of reaping, or 3. Confirm that persistent Monitor is the intended primitive for long-lived supervision, so tooling conventions can standardize on it. Environment: macOS (Darwin 24.6), interactive session (not -p mode).

Environment Info

• Platform: darwin
• Terminal: iTerm.app
• Version: 2.1.212
• Feedback ID: 06c0cf3f-f9bd-4550-b074-5bf2b53b6f65

Errors

[]