Leon-Drq/openagentskill

What it solves

OpenAgentSkill addresses the difficulty of finding, evaluating, and safely installing reusable "skills" (third-party code/tools) for AI agents. It prevents agents from blindly installing popular but unsuitable or risky repositories by providing a decision layer that audits skills and matches them to specific tasks.

How it works

The system acts as a resolver between an agent's task and the execution of a skill. When an agent needs to perform a task, it calls the Resolve API to find the best-fitting skill based on task relevance, trust scores, maintenance history, and license clarity. The process follows a loop: the agent requests a skill, receives a recommended install plan and risk level, runs the skill in a sandbox, and then reports the outcome back to the system to improve future rankings.

Who it’s for

• AI Agent Developers: Those building agents (like Codex, Claude Code, or Cursor) that need a reliable way to extend their capabilities with third-party tools.
• Skill Authors: Developers who want to make their open-source tools discoverable and verifiable for AI agents.
• End Users: People using agents who want an inspectable and audited trail of what tools their agents are installing.

Highlights

• Task-to-Skill Resolver: An API and CLI that recommends the best skill and ranked alternatives for a specific task.
• Trust Score v5: A decision-support signal based on repository evidence, install safety, and real-world agent outcomes.
• Agent Integration Kit: Ready-to-use templates for integrating the resolver into popular agent frameworks.
• Machine-Readable Surfaces: Provides data via JSON, OpenAPI, and specialized manifests (llms.txt) for seamless agent consumption.