executing

Per-agent execution loops are automated by coordinating initialization, bead selection, file reservation, implementation, verification, and reporting. This ensures deterministic bead progression and safe pause/resume behavior across swarm-based workers.

A bead loop is a structured execution cycle for per-agent tasks that enforces deterministic progression, explicit verification steps, and auditable completion records. It manages versioned beads, context decisions, and standard history artifacts.

Safe pause and resume semantics are handled natively during bead execution by enforcing guarded handoffs that preserve locked decisions and maintaining standard history evidence artifacts. This ensures verifiable completion records across interruptions.

Standalone single-agent bead execution is supported as a degraded worker mode triggered when preflight advises against swarming. It performs the same initialization, implementation, verification, and reporting loops without swarm coordination.

Versioned bead completion is verified and reported by enforcing the bead structure described in SKILL.md and generating standard history evidence artifacts. This approach provides explicit verification steps and auditable records for each executed bead.

File reservation in a bead execution loop requires identifying the next bead from the live graph and reserving its target files before implementation. This ensures safe coordination and guarded handoffs across concurrent swarm workers.