Build your agent from 200,000+ skills via skill RETRIEVAL & ORCHESTRATION. AgentSkillOS addresses the challenge of discovering and composing skills from a rapidly growing ecosystem. It targets engineers and researchers building complex AI agents, enabling them to efficiently discover, orchestrate, and run skill pipelines for diverse tasks, thereby enhancing agent capabilities and productivity.

How It Works

AgentSkillOS employs a novel "skill tree" to hierarchically organize over 200,000 skills, facilitating creative and structured discovery beyond simple semantic search. Skill retrieval automatically selects task-relevant skills, which are then composed into a Directed Acyclic Graph (DAG) for orchestration. This approach manages execution order, dependencies, and data flow, allowing complex tasks to be solved by multi-skill pipelines. A GUI supports human-in-the-loop intervention, while a modular architecture allows pluggable retrieval and orchestration components.

Quick Start & Requirements

• Prerequisites: Python 3.10+, Claude Code (in PATH), cc-switch (for alternative LLMs).
• Installation: Clone the repository, cd AgentSkillOS, pip install -e ., copy and edit .env with API keys.
• Run: python run.py --port 8765 for the web UI.
• Pre-built Skill Trees: Downloadable sets ranging from ~50 to ~10,000 skills are available via Google Drive and Baidu Pan.
• Configuration: API keys and endpoints for LLM and embedding models are configured in the .env file.
• Documentation: Detailed workflows are available on the project's landing page.

Highlighted Details

• Skill Tree Discovery: Hierarchical organization of skills for efficient and creative task-relevant discovery.
• DAG Orchestration: Composes skills into workflows, managing dependencies and data flow.
• GUI: Provides human-in-the-loop control, auditing, and steering capabilities.
• Curated Skill Pool: Features a collection of high-quality skills selected based on implementation, stars, and downloads.
• Benchmark: Evaluates performance on 30 multi-format creative tasks using pairwise comparison and Bradley-Terry aggregation.
• Batch CLI: Enables headless, parallel execution of tasks via YAML configurations with resume support.
• Observability: Detailed logs and metadata for faster debugging and iteration.

Maintenance & Community

Recent news items from March 2026 indicate active development, including the release of a new project homepage, benchmark, modular architecture, and batch CLI. No specific community links (e.g., Discord, Slack) or details on core contributors/sponsorships are provided in the README.

Licensing & Compatibility

The provided README does not specify a software license. This omission requires clarification regarding usage restrictions, particularly for commercial applications or integration into closed-source projects.

Limitations & Caveats

The project lists several areas for future work, including interactive agent execution, plan refinement, and auto skill import, suggesting these features are not yet implemented. The system has a dependency on Claude Code and requires configuration of external LLM and embedding API keys.