Boundary
Task state now lives outside the transient agent loop. That makes dependencies, blocking, and completion visible to the whole system.
What this adds
Make work durable.
This lab stores tasks, statuses, and dependencies in a durable file. The key feature is not storage by itself. It is ready-task detection: what can be worked on right now?
Once this exists, a restarted process or second worker can recover the state of the job instead of starting from memory.
Done when
A restarted process can reload the graph and identify the next unblocked task without recomputing the world from scratch.
Real-world analog
Taskwarrior is a useful everyday comparison: durable tasks, explicit state transitions, and clear next-action visibility. AI-near tools like Beads live nearby.
Why this matters
Durable state is where agent work stops feeling like a chat session and starts feeling like a system.