convez.ai
MCP ServerAutomaçãoAvançado
ThanhNguyxnOrg

SkillGauge

ThanhNguyxnOrg/SkillGauge

Auditoria científica e ranking de skills para agentes IA com 100 critérios

Ver no GitHub
BomPluginScore · de 0 a 100

Motor de auditoria estática e plataforma de leaderboard para instruções de processos agênticos, utilizando uma matriz científica rigorosa de 100 critérios. Avalia qualidade estrutural, segurança e eficiência operacional de skills de IA sem necessidade de execução dinâmica.

Instalação

$npx convez-ai ThanhNguyxnOrg/SkillGauge

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
87%
Último commit
há 1 dia
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Validar qualidade e segurança de instruções de agentes IA antes da produção
  • Rankear e comparar skills de diferentes repositórios em leaderboard público
  • Automatizar verificação de conformidade de process instructions via CI/CD

Veredito

Ferramenta niche mas bem-engenhosada para avaliação estática de skills agênticos. Oferece metodologia científica robusta (7 dimensões, 100 critérios) e integração GitHub elegante, mas o mercado-alvo é restrito a equipes que trabalham intensamente com agentes IA estruturados.

README

A tradução para o português deste README ainda está na fila.

🏆 SkillGauge: Agent Skills Leaderboard & Scientific Auditor

SkillGauge is an automated static auditing engine and leaderboard platform for agentic process instructions (such as Anthropic-style SKILL.md or .agent/skills/ files).

Instead of evaluating AI agent executions dynamically (which is expensive and non-deterministic), SkillGauge audits the structural quality, safety, and operational economy of skill prompts directly through a strict 100-criteria scientific matrix.


⭐️ Top Agent Skills Leaderboard

The live ranking table is dynamically managed and updated on our web dashboard:

👉 View the Live Leaderboard


📖 Documentation Directory

To learn more about the scientific methodology, CLI commands, and automated pipelines, please check the detailed guides:

  1. Scientific Specification Detailed formulas, quality tier divisions, and academic whitepaper references for each of the 100 criteria across the 7 dimensions:

    • Dim A: Instruction Quality & Clarity
    • Dim B: Context & Memory Management
    • Dim C: Safety, Alignment & Security
    • Dim D: Tool-Use & MCP Clarity
    • Dim E: Robustness & Exception Handling
    • Dim F: Operational & Inference Economy
    • Dim G: Syntax, Structure & Metadata
  2. CLI Usage Guide How to install, compile, and run the CLI for local files, directories, or remote Git targets using either HTTPS links or owner/repo shorthand notation.

  3. GitHub Actions CI Bot Detailed guide on how the modular, split GitHub Actions workflows (submit-repo-bot.yml, pr-verification.yml, update-leaderboard.yml, deploy-pages.yml) orchestrate automated scan-and-submit tasks and post-merge optimizations.


🚀 How to Submit a Skill

Submitting your agent skills to the global leaderboard is incredibly simple and requires no login:

  1. Via Web Dashboard (Recommended):

    • Navigate to the Submit Skill tab on the live dashboard.
    • Paste your GitHub repository URL (e.g., https://github.com/owner/my-agent-skills).
    • Click Submit. The backend will trigger a background Action that clones your repository, audits all valid skills, and opens a Pull Request on your behalf.
    • Follow the live Pull Request link returned by the website to view your detailed score report!
  2. Via Manual Pull Request:

    • Place your markdown skill files (which must start with YAML frontmatter containing name: and description:) in a folder named skills/<your-repo-name>/<skill-name>.md.
    • Open a Pull Request against the main branch.
    • Our verification runner will automatically check for duplicate entries, score your skill against the 100-criteria scientific matrix, and comment with a detailed report card.

📁 Repository Structure

skills/
├── [repository-name]/
│   ├── [skill-name].md
│   └── SKILL.md
  • Repository Detection: The CLI parses the repository name from the subdirectory (e.g. skills/superpowers/brainstorm.md -> repository superpowers). For Git targets, the repo name is extracted directly from the repository URL.
  • Leaderboard Name Format: Entry names are rendered as clean skill names (e.g. brainstorm), and their origin repositories are displayed in the dedicated "Repository" column.
Ver o original no GitHub

Bom para

Relacionados