Long-Running
Domain: Automation & Scheduling
Motivation
Long-running work is the reason the job system exists: it must stay alive while the agent’s turn ends and later turns resume. The job registry bounds it with per-job output rings and pump polling (pumpPollMs, default 150 ms), and workflow runs have no overall elapsed deadline, only per-backend execution limits.
Concrete Example
A test suite or a build started as a background job keeps streaming stdout into its ring; the model reads it later with job_output, and updateProgress can publish a live progress line into every projection.
Analogy
A marathon, not a sprint: the system’s job is to keep the runner supplied and counted, not to hurry them.