Lifecycle

Runs need recoverable state.

The lifecycle is designed around fresh, resume, halt and done states. The point is not to hide failure; it is to keep enough state for a human or tool to decide what should happen next.

PUBLIC ALPHA Route status
run / protocol/run-lifecycle gated
plan accepted
validate_plan accepted
implement evidence open
review_changes findings open
repair_changes queued
final_acceptance awaiting signoff

Resume

Continuation should not rebuild context.

A paused or interrupted run can resume from its checkpoint, preserving the plan, events and evidence accumulated so far.

Halt

Stopping is a first-class outcome.

A halted run should leave a clear reason and enough artifacts for the operator to inspect before deciding on a new attempt.