EvoSkill – Automated skill discovery for multi-agent systems

Summary: EvoSkill is an open-source framework that identifies gaps in coding agents' capabilities and synthesizes reusable skills from failed trajectories to enhance performance on long-horizon tasks. It improves results on complex benchmarks and supports skill generalization across tasks.

What it does

It detects deficiencies in agent operations and generates new skills through an evolutionary loop using failed attempts, improving agent performance on challenging benchmarks.

Who it's for

Developers and researchers working on multi-agent systems and coding agents for complex, long-horizon tasks.

Why it matters

It addresses capability gaps in agents by creating reusable skills that boost performance and generalizability across different benchmarks.