🌟 Why SkillsVote?

Say goodbye to massive, hardcoded, and bloated skill lists! SkillsVote is building an intelligent, dynamic ecosystem for skill recommendation, feedback, and long-term skill evolution.

Acting as a smart gateway, SkillsVote delivers just-in-time recommendations, dynamically routing your AI agents to the exact skills they need. The result? ⚡ Maximized token efficiency and 🎯 sky-high task success rates.

🌍 The World's Largest Skill Library

At the product level, we are mining the vast open-source universe of GitHub to build an unprecedented library:

• 🔥 1.68M+ discovered SKILL.md files
• 💎 790K+ format-valid skills after validation

💡 About This Repository

This is the open-source, local-first core of SkillsVote. It equips you with:

1. a powerful static analysis pipeline for skill profiling, and
2. a smart local agentic recommendation pipeline.

📰 Latest News

• 🌅 [2026-04-09] Special Share. Core contributor's share on Linux.do.
• 📣 [2026-04-08] Social Launch. Our launch announcement is now live on WeChat Blog and rednote.
• 🚀 [2026-04-03] Launch Day! Published the very first open-source release of our recommendation and evaluation demos.

✨ Key Features

1. 🔍 Rich Skill Profiling. SkillsVote doesn't just read skills; it understands them. We build a structured, comprehensive profile for every skill—covering OS requirements, env variables, CLI needs, and MCP dependencies. This makes browsing our website a breeze while ensuring strict quality control.
2. 🏗️ Real Task Construction & Execution. We go far beyond static inspection! Skills that pass our verifiability screening are put to the test: we construct executable tasks, controlled sandboxes, and strict validators to prove that a skill actually helps agents get real work done.
3. 🧠 Agentic Recommendation Engine. Given a user task and a local skill directory, our agentic navigation searches your directory and returns the perfect recommended skill set, complete with foolproof usage guidance.

For a detailed breakdown of the quality and verifiability criteria used in our evaluation pipeline, see Appendix: Evaluation Metrics Unpacked.

🚀 Quick Start

Option 1: Install the Hosted Skill (Recommended)

🤖 Agent Setup Prompt

Supercharge your agents (Codex, Claude Code, OpenClaw) by integrating SkillsVote directly! Just drop this prompt into your agent:

1. Install the skill by running `npx skills add MemTensor/skills-vote --skill skills-vote`
2. Create or update `.env` file located in the root directory of the installed `skills-vote` skill and set `SKILLS_VOTE_API_KEY="YOUR_API_KEY"`

Do not configure this as a system or user-level environment variable unless explicitly requested.

🔧 Manual Setup Alternative

Are you a CLI warrior? Set it up manually based on your OS:

Windows PowerShell

[Environment]::SetEnvironmentVariable("SKILLS_VOTE_API_KEY", "YOUR_API_KEY", "User")
npx skills add MemTensor/skills-vote --skill skills-vote

MacOS/linux (Bash/Zsh)

# For zsh, use ~/.zshrc instead
echo 'export SKILLS_VOTE_API_KEY="YOUR_API_KEY"' >> ~/.bashrc && source ~/.bashrc
npx skills add MemTensor/skills-vote --skill skills-vote

[!note]
Don't forget to replace YOUR_API_KEY with your actual key!

Option 2: Run the Local Demo 🏠

Want to test drive the core engine locally? Just follow these 3 easy steps:

1. Install dependencies

uv sync

2. Configure Environment

Copy the example config and fill in your Anthropic credentials.

cp .env.example .env

Use ANTHROPIC_API_KEY when calling the official Anthropic API. Use ANTHROPIC_AUTH_TOKEN and ANTHROPIC_BASE_URL when calling a third-party Anthropic-compatible service.

3. Run the examples

bash examples/evaluate.sh
bash examples/recommend.sh

Outputs are written to output/evaluate_results.jsonl and output/recommend_result.json.

You can override the query with:

bash examples/recommend.sh -q "Summarize a pull request and highlight risky changes"

If you want to use your own local skills, update skills_dir in scripts/configs/recommend.yaml and scripts/configs/evaluate.yaml, then rerun the same commands.

📎 Appendix

📊 Evaluation Metrics Unpacked

Table 1. Quality Evaluation

Metric	Description	Why it matters
Content Consistency	Whether the skill stays centered on one clear, stable purpose and whether the rest of the content consistently supports that purpose.	A recommended skill should be a stable capability unit, not a mixed bundle of unrelated topics.
Reference Completeness	Whether the referenced scripts, resources, templates, and dependencies are present and usable as documented.	Broken references and missing artifacts are one of the most common failure modes in open-source skill libraries.
Task Orientation	Whether the skill provides actionable guidance for completing work rather than only background information.	SkillsVote is recommending executable skills, not just retrieving knowledge.

Table 2. Verifiability Evaluation

Metric	Description	Why it matters
Success Verifiability	Whether results can be judged programmatically with low ambiguity.	Subjective skills such as brainstorming or poetry writing are not suitable for automatic validation.
Environment Controllability	Whether the required environment can be reproduced, reset, and executed reliably in a controlled sandbox.	Skills that depend on live external systems or open-world state are hard to benchmark deterministically.
Task Constructability	Whether many realistic task instances and validators can be generated at reasonable cost.	Some domains require expensive hardware, large datasets, or heavy manual work and do not scale well for evaluation.

📄 License

This repository is licensed under the MIT License. See LICENSE.