A deploy killed my OTP processes but not the Docker containers running my coding agents, so the board showed a task “executing” for 18 hours after its PR was already merged.

The fix: a reconciler GenServer on a 60s tick, container adoption through an exit-marker file, and a pure decide/4 — the incident is now a unit test with the real timestamps.

https://t0ha.dev/en/elixir/devops/2026/09/04/supervising-what-you-dont-own.html