anbeime/skill: A Self-Updating AI Agent Skill Library and Validator

What anbeime/skill Is and Who It Is For

anbeime/skill is a repository that serves two distinct purposes at once. First, it auto-crawls the awesome-agent-skills index on GitHub every 24 hours and stores those results in data/skills.json, giving users a local mirror of the public skill ecosystem that stays current without manual updates. Second, it curates and packages 63 original Chinese-language skills covering content creation, video production, e-commerce, programming, finance, and several other verticals, stored in the skills/ directory alongside zip backups.

The repository is aimed at AI agent users who want a single place to browse, validate, and install skills for runtimes like Claude Code, Cursor, and Codex, and at developers building Chinese-language automation workflows who need domain-specific skill packs not available in the upstream index.

The Dual-Library Design: Official and Local Skills

The README divides the skill inventory into two categories. Official skills total 182, sourced from the awesome-agent-skills list and attributed to teams including Anthropic (17 skills covering docx, pptx, xlsx, pdf, mcp-builder, and webapp-testing), Vercel (7 skills), Cloudflare (7 skills), Trail of Bits (26 security skills), Hugging Face (8 skills), Google Labs (2 skills), Stripe (2 skills), Supabase, Expo, and Sentry. Each entry is stored with its name, description, GitHub link, category, source repository, and crawl timestamp.

Local skills total 63, organized into 11 subcategories. Notable clusters include 10 content creation skills (covering article illustration, WeChat publishing, Xiaohongshu image generation), 9 video creation skills, 7 e-commerce and marketing skills, and a full-pipeline e-commerce skill (ecommerce-full-pipeline) described as covering product sourcing, listing, promotion, video, fulfillment, and resale selection. Six skills are marked as system-built-in: pptx, xlsx, pdf, docx, find-skill, and skill-creator.

Installing and Running the Skill Library

Clone the repository and install the Python dependencies:

git clone https://github.com/anbeime/skill.git
cd skill

pip install -r requirements.txt

requirements.txt lists requests, beautifulsoup4, schedule, and lxml as core dependencies. Flask and flask-cors are listed as optional, for the API server example. After install, run a one-off crawl update or start the daemon:

python main.py --once
python main.py --daemon

The daemon runs the crawl on a 24-hour schedule automatically. Additional flags include --stats to display a skill count summary, --export skills.csv to dump the official skill list to a CSV file, and -v for verbose log output during a crawl. The crawled data lands in data/skills.json and the local skill metadata in data/local_skills.json, both readable directly:

cat data/skills.json
cat data/local_skills.json

The SKILL.md Validator and Tooling

The repository ships a standalone validator in tools/skill_validator.py that checks whether a SKILL.md file conforms to the Agent Skills protocol specification. Four operations are supported:

python tools/skill_validator.py validate
python tools/skill_validator.py validate skills/agent-team
python tools/skill_validator.py read-properties skills/agent-team --json
python tools/skill_validator.py to-prompt skills/agent-team

The first command validates all skills in the directory. The second validates a single named skill. read-properties extracts the YAML frontmatter from a SKILL.md and returns it as JSON. to-prompt generates an XML-formatted prompt representation of the skill for direct use in an AI agent call.

The docs/ directory contains the Agent Skills specification document (docs/specification.md), a best-practices guide for writing high-quality skills (docs/best-practices.md), a quickstart tutorial claiming 30-second skill creation (docs/quickstart.md), and migration and cleanup guides for managing local skill collections.

Limitations: No License, Windows Paths, and Coverage Gaps

The repository declares no license. The LICENSE field in the repository metadata is listed as unknown, meaning there is no documented permission to redistribute or modify the contents. This is a practical problem for organizations that need a clear legal basis before using the local skills in production or including them in a distributed product.

The README's backup documentation references local Windows paths (D:\tool\skills\) suggesting the primary author's development environment is Windows. The backup format (71 zip files) is documented but the zip files themselves are stored locally, not in the repository. Users on other platforms or seeking a portable backup solution need to adapt the described process.

The local skill collection is heavily weighted toward Chinese-language content creation and social media workflows on Chinese platforms (WeChat, Xiaohongshu, Douyin). Engineers working in other languages or on other platforms will find the upstream official skills from Anthropic, Vercel, and Cloudflare more directly applicable, and those are already available via the awesome-agent-skills list without requiring this repository.

Alternative: awesome-agent-skills

The VoltAgent/awesome-agent-skills repository is the upstream source that anbeime/skill crawls. It is a curated list of AI agent skills maintained manually, without the automation layer. The difference is scope and tooling: awesome-agent-skills is a Markdown index for browsing, while anbeime/skill adds auto-crawling, local storage in JSON and CSV, Chinese-language local skills, the SKILL.md validator, and the Flask-based API example.

For users who only need to discover existing published skills, the awesome-agent-skills list is sufficient and requires no local install. anbeime/skill is the better fit when the goal is to run automated updates, export skill data for analysis, validate custom SKILL.md files, or access the curated Chinese-language skills that are not in the upstream list. The two repositories are complementary rather than competing: awesome-agent-skills is the authoritative upstream, and anbeime/skill is a consumer and mirror that adds domain-specific local packs and tooling around it.

Data Format and Automation Architecture

The crawler (crawler.py) fetches the awesome-agent-skills repository, parses skill entries, and writes them to data/skills.json. Each entry in the JSON carries name, description, link, category, source, and crawled_at fields. The scheduler (scheduler.py) wraps the crawler in a schedule loop for daemon mode. data_manager.py handles the local skill metadata in data/local_skills.json, which uses a different schema: metadata fields cover total_skills, core_skills, sub_skills, and system_builtin counts.

The last push was on 2026-09-26. No license is documented in the repository, which limits redistribution rights and creates ambiguity for organizations that need legal clarity. The online skill store at skill.miyucaicai.cn is listed as the project's homepage and provides a web interface for browsing the indexed skills without cloning or running the local code. The Vercel deployment configuration (vercel.json) in the repository powers this hosted web interface.

Editorial conclusion

anbeime/skill is a practical starting point for anyone who wants a local mirror and index of available AI agent skills, or who wants pre-packaged Chinese-language skills for domains like e-commerce automation, PPT generation, or video production. The repository carries no license declaration, which creates legal ambiguity for organizations that need clear redistribution rights. Developers integrating the skills into production workflows should verify each skill's own license before deploying, since the repo aggregates skills from multiple upstream authors.

Frequently asked questions

Official sources

1. anbeime/skill on GitHub
2. Issues
3. Project website
4. README