checkout

You prevent duplicate work across multiple agents by using atomic task locking. This mechanism creates a unique lock per task, ensuring only one agent can process it at a time and eliminating wasted compute.

Locked tasks release automatically on agent death or timeout to avoid stuck tasks. You can also manually manage locks using status, list, and release commands to reassign work when needed.

You check the status of locked tasks using built-in status and list commands. These management features let you view active locks, check task availability, and manually release tasks to reassign work.

Yes, this atomic task locking works with filesystem-based locks for autonomous systems. It implements race-free locking directly on the filesystem, making it suitable for multi-agent coordination without external dependencies.

Filesystem-based locking relies on the local filesystem, meaning limitations arise with distributed agents on separate machines unless a shared filesystem is mounted. Auto-release mitigates stuck tasks but requires timeout configurations.