libukai/awesome-agent-skills

Chinese Simplified | English | Japanese

This project is committed to following the principle of ‘few but refined,’ collecting and sharing the highest quality Skill resources, tutorials, and practical cases to help more people easily take the first step in building personalized agents.

If you find this project helpful, please give 🌟 it a click so more people can know about it. Also, feel free to follow my X account @李不凯正在研究 for the latest resources and practical tutorials on Agent Skill!

Quick Start

Skill is a lightweight agent construction solution that encapsulates specific business processes and industry knowledge, enhancing AI’s professional capabilities in performing specific tasks.

When facing repetitive task demands, you don’t need to repeatedly enter background information in every conversation. Simply install the corresponding Skill, and the Agent can acquire specialized skills in the field.

After nearly a year of iterative evolution, Skill has become the standard solution for enhancing AI vertical capabilities, gaining broad support from mainstream Harness frameworks and AI products.

Support status

The Skill Open Specification has been adopted by many hosts such as Claude Code, ChatGPT and Codex, GitHub Copilot, Cursor, Gemini CLI, VS Code, OpenCode, Kiro, JetBrains Junie, and others. Search paths and experimental field support may vary among hosts; please refer to the Agent Skills Client Showcase and corresponding product documentation for reference.

Standard Structure

Agent Skills is an open specification initiated by Anthropic and maintained by the community. Each Skill is a standardized folder containing processes, data, scripts, and other resources; Agents reduce irrelevant context through progressive loading.

In the most streamlined skill, at least the file SKILL.md must be included.

The YAML frontmatter for SKILL.md must include name and description, and can also declare license, compatibility, metadata, and experimental allowed-tools.

name The maximum length is 64 characters, and only lowercase letters, numbers, and hyphens can be used, and must match the parent directory; description The maximum length is 1024 characters, requiring both explanations of “what it is doing” and “when to use.” The main text is recommended to be under 500 lines and 5000 tokens; detailed content should be split into a separate resource file.

In addition to SKILL.md, you can also add the following files or directories as needed:

• scripts/: Stores directly executable code, suitable for solidifying operations that are repetitive, complex, or require deterministic results, such as data processing, format conversion, and result verification.
• references/: Stores supplementary materials that the Agent only needs to read for specific tasks, such as domain knowledge, technical documentation, examples, and data format descriptions.