See what a run is doing while it runs

After this entry you'll know how to answer "is this run alive, and what is it doing?" without digging through logs. Until now a long-running step was a black box: the run said Working and nothing more until the step finished. Now the run view shows live activity as it happens.

What changes for you

• Live activity in the timeline. Every action the agent takes (running a command, editing a file, searching the repo) appears in the run's timeline as it starts, with a short one-line headline. Headlines are scrubbed of anything sensitive before they leave the agent.
• A heartbeat on every run. The Working card now carries a live line showing the last action, how fresh it is, and spend so far, and the fleet view shows last-activity age and current phase for every run in flight.
• Stall detection. If a running job goes silent for more than five minutes it is flagged as stalled: the run's timeline records it, the fleet view shows it, and one alert goes to the ops channel. When activity resumes the flag clears and the recovery is recorded too, so you get one alert per genuine episode, not a page every minute.

None of this can break a run: if the observability plumbing fails, the run carries on and you simply see less.