github-picks-skill

An AI agent skill that delivers daily GitHub project recommendations — like Steam's daily game picks, but for open-source repos.

What it does

Every day, this skill searches GitHub for interesting projects and picks 5 for you:

• Fresh discoveries: Prioritizes hidden gems (< 10k stars) over mega-repos everyone already knows
• Smart dedup: Never recommends the same repo twice
• Preference learning: Tell it what you liked/disliked, it adapts future recommendations
• Quality filters: Excludes archived, stale (> 90 days inactive), and no-description repos

Output is a Chinese Markdown report with a summary table and detailed project descriptions.

Structure

SKILL.md                 # Skill instructions (the prompt)
scripts/data_manager.py  # CLI for state management (status/record/feedback)
data/                    # User data (history, preferences, dedup list)
references/schemas.md    # JSON schemas for data files
evals/evals.json         # Evaluation test cases

Usage

This skill is designed for AI coding agents that support the Anthropic Skills format. Install it as a skill in your agent, then ask:

"推荐今天的 GitHub 项目"
"今日 GitHub picks"
"有什么有趣的新开源项目"

After receiving recommendations, give feedback:

"喜欢 autokernel 和 OpenBrand，不喜欢教育类的"

The skill will persist your preferences and use them for future recommendations.

Requirements

• gh CLI (authenticated with GitHub)
• Python 3.8+
• pyyaml (optional, for some edge cases)

License

MIT