agent-topology-optimizer

Analyze the current topology's latency, throughput, and bottleneck metrics, then evaluate candidate structures like hierarchical, mesh, or hybrid against the workload profile. Apply a migration plan only when the projected improvement exceeds the configured threshold.

The Skill implements genetic algorithms, simulated annealing, particle swarm optimization, graph partitioning, and ML-based placement. It runs multiple algorithms in parallel and combines results through ensemble optimization to select the best placement.

Protocol selection depends on the communication pattern between agent pairs. UDP and MQTT offer low latency and overhead, gRPC balances low latency with high reliability, while TCP prioritizes reliability at higher overhead.

Avoid reconfiguration when the evaluated improvement falls below the minimum threshold (default 0.1) or when migration downtime outweighs benefits. The optimizer returns no recommendation when no significant improvement is found.

Yes, it uses MCP hooks such as swarm_status, performance_report, bottleneck_analyze, topology_optimize, and swarm_scale to gather metrics, apply optimizations, and monitor impact after changes.