One `rm -f` Silently Disabled My Crash Detection

A developer working on the open-source tool agentwatch discovered a critical bug where a cleanup script was silently disabling crash detection. The daemon relied on a PID file to determine if a restart followed a crash, allowing it to synthesize a 'recording-gap' record. However, the test suite's cleanup process was deleting this PID file, causing the daemon to incorrectly assume it was performing a fresh start. Because the daemon simply skipped the recovery logic without throwing an error, the failure remained invisible until explicit assertions were added to the test suite. The author highlights the danger of using files as state signals in automated environments, noting that cleanup routines often fail to distinguish between 'stale' files and essential system markers. The issue was resolved by updating the test harness to preserve the PID file, ensuring the recovery path is correctly exercised during testing.