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.

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

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.

• 🌅 [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.

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.

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.

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

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.

Table 1. Quality Evaluation

Table 2. Verifiability Evaluation

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