agentskills: Explore Resources for Building LLM Agent Skills

Overview

scienceaix/agentskills is an annotated resource directory for people exploring how to build and evaluate skills for large language models. It connects the emerging agent-skills concept with adjacent areas such as MCP, tool use, GUI agents, benchmarks, and agent frameworks.

This is a discovery and reference project, not a software library or a collection of ready-to-install skills. It is useful when you want a broad starting point for research or implementation choices, and less suited to readers looking for a single prescribed stack or step-by-step course.

Key Features

• Curates official Agent Skills announcements, documentation, repositories, and courses.
• Groups MCP specifications, SDKs, repositories, and community resources together.
• Includes research papers on skill learning, composition, tool use, GUI agents, and related topics.
• Lists open-source agent, browser-automation, computer-use, and coding projects.
• Provides links to agent, function-calling, GUI, and web-browsing benchmarks and leaderboards.
• Adds tutorials, related curated lists, a project timeline, and a citation for the associated survey.

Use Cases

• Researchers can use the categorized papers and benchmarks to orient a literature review on LLM skills or agents.
• Developers deciding how to extend an LLM with reusable skills can compare documentation, frameworks, and implementation examples.
• Teams evaluating tool-use or computer-use approaches can find relevant projects and benchmark references in one directory.
• Educators and learners can use the tutorials and courses as a starting point for studying agent architecture, skills, and MCP.

Getting Started

Browse the README and follow the section links to the resources relevant to your topic. The repository is a curated list and does not document an installation command.

Alternatives

• awesome-agent-skills: Indexes reusable agent skills for coding assistants, while agentskills surveys skills alongside papers, tools, frameworks, and learning resources.
• Awesome-AI-Agents: Focuses on autonomous and multi-agent projects and benchmarks, while agentskills covers skills, MCP, tool use, and computer-use research.

Considerations

• The project provides links and annotations rather than executable software, so it does not offer a unified setup, API, or implementation workflow.
• Its coverage spans a wide range of related areas. Check individual linked sources for current details, requirements, and claims.
• The repository was created in 2026 and is actively updated according to the supplied repository metadata; a broad, fast-moving resource list may change as the ecosystem evolves.
• The README includes many dated resources and performance claims. Treat those as context from the linked sources, not as independently verified results from this repository.