Agent Swarm

The Agent Swarm skill extends the Multi-Agent skill with swarm intelligence patterns, enabling large numbers of lightweight agent instances to collaborate on problems that benefit from parallel exploration and collective decision-making. Unlike Multi-Agent's structured orchestrator-worker model, Agent Swarm operates with emergent coordination — agents share a common goal, a shared knowledge pool, and a voting or consensus mechanism for resolving disagreements. Typical swarm sizes range from 5 to 50 agents depending on task complexity. Use cases include parallel web research where many agents explore different sources simultaneously and merge findings, competitive analysis where each agent evaluates a different competitor, or code review where multiple agents independently audit different aspects of a codebase. A blackboard architecture stores intermediate results accessible to all swarm members, preventing duplicated work. Swarm progress is visualized as a coverage map showing which sub-problems have been addressed and which remain open. Cost controls cap total token spend for the swarm and shut it down gracefully when limits are reached, returning whatever results have been gathered so far.

Installation

clawhub install agent-swarm

Install: clawhub install agent-swarm