Skill Compose is an open-source platform for building and running skill-powered AI agents, designed to eliminate complex workflow graphs and CLIs. It targets engineers and researchers by offering a streamlined approach to agent development, enabling users to describe desired agent behavior and have the platform automatically compose agents from reusable, versioned skills. This facilitates rapid prototyping and deployment of sophisticated AI agents.

How It Works

The core innovation lies in treating "skills" as first-class, versioned artifacts—encompassing contracts, references, rubrics, and helpers—rather than brittle, imperative graphs. The "Skill-Compose My Agent" workflow allows users to describe their agent's purpose, whereupon the platform intelligently finds existing skills, drafts necessary new ones, and wires them together. This declarative approach, coupled with tool and MCP wiring without manual glue code, significantly simplifies agent composition and management.

Quick Start & Requirements

Get started quickly using Docker. Clone the repository, navigate to the docker directory, copy the example environment file (.env.example to .env), and start the services with docker compose up -d. A default LLM API key (e.g., MOONSHOT_API_KEY) is required, which can be configured in the .env file or the Web UI post-launch. Access the UI at http://localhost:62600. Full documentation and API references are available online.

Highlighted Details

• Skills as First-Class Artifacts: Versioned, reviewable skill packages with defined contracts and rubrics.
• Automated Agent Composition: The "Skill-Compose My Agent" workflow automatically finds, drafts, and integrates skills based on user descriptions.
• Container-First Isolation: Agents run in isolated containers or K8s pods for reproducible and clean execution environments.
• Comprehensive Skill Lifecycle Management: Supports GitHub import, one-click updates, multi-format import/export, version history, diffing, and local synchronization.
• Skill Evolution: Agents and skills can be improved automatically using feedback and execution traces, with reviewable proposed rewrites.
• Instant Publishing: Agents can be deployed with one click as shareable Web Chat links or integration-ready API endpoints.

Maintenance & Community

The provided README does not detail specific contributors, sponsorships, or community channels like Discord or Slack.

Licensing & Compatibility

The project is licensed under the Apache License 2.0. This license is generally permissive and compatible with commercial use and closed-source linking.

Limitations & Caveats

Some features described or shown in the README may still be in development. The platform requires configuration of at least one Large Language Model (LLM) API key to function.