EvoSkill Launches Open-Source Framework That Turns AI Agent Failures Into Reusable Skills Across Multiple Models

EvoSkill launches an open-source framework that transforms AI agent failures into reusable skills, running a self-improvement loop across multiple coding agents and model providers — with skills transferable across different agents, models, and tasks.

• EvoSkill is an open-source framework that automatically discovers and synthesizes reusable agent skills from failed trajectories, running a self-improvement loop that analyzes failures, proposes skill or prompt changes, and tracks top-performing agent configurations as git branches.
• The framework supports multiple coding agents including Claude Code, Codex CLI, OpenCode, OpenHands, Goose, and Harbor, and is compatible with major model providers such as Anthropic, OpenAI, OpenRouter, and Fireworks AI.
• EvoSkill offers cross-agent, cross-model, and cross-task skill transferability, meaning skills evolved with one agent or model can improve performance on entirely different agents, models, or tasks.