What problem does it solve? Coordinating autonomous worker agents across multiple hosts leads to duplicated work, stale ledger reads, and unclaimed tasks. This Skill provides a recurring execution loop that lets any agent discover, claim, implement, checkpoint, and complete shaped work packets from a shared plan ledger without colliding with other hosts. ## Core Features & Use Cases - Batch Work Selection: Runs a minimax-scored batch selector that groups eligible packets by epic and release target, preventing priority-first starvation and cross-host collisions. - Lease-Based Claiming: Flips packet status to in_progress and pushes the claim before work begins, with cross-branch claim checks and timestamp-based collision arbitration. - Cycle Safety Guards: Enforces checkout locks, committable-branch guards, preflight instrument checks, and startup boundary snapshots so cycles are attested and never overlap. - Use Case: A scheduled worker agent on a Linux host wakes up, acquires the checkout lock, selects a budgeted batch of release-targeted packets, claims them with a pushed status change, implements the work, and finalizes the cycle with a verified commit. ## Quick Start Ask the agent to run an advance-work-from-plan cycle to select, claim, and complete the next eligible work packet for this host.
