What problem does it solve? When multiple AI agents work in parallel on a shared goal, they need a way to claim tasks without duplication and exchange findings mid-flight without a central server. This Skill provides a file-based coordination protocol under ~/.cache/agents-supervisor so peer agents can claim work and message each other using only atomic file renames. ## Core Features & Use Cases - Atomic Task Claiming: Agents list pending tasks and claim them by renaming files from --pending to --claimed--<agent>, preventing double-assignment through filesystem atomicity. - Peer Inbox Messaging: Agents send short messages by writing new files to inbox/to-<peer>--from-<you>--<slug>.md, never editing another agent's files. - Shared Knowledge Space: Long findings and consensus go into a shared/ directory so messages stay short and reference paths instead. - Use Case: A lead agent spawns three workers to debate an authentication design. Each worker claims sub-tasks from the team's tasks/ directory, exchanges critiques via inbox files, and writes the final consensus to shared/ before reporting back to the lead. ## Quick Start Set up a team mailbox under ~/.cache/agents-supervisor for my worker agents so they can claim tasks and message each other during this multi-agent run.