skills  by mizchi

Agent skills framework for AI development and analysis

Created 1 month ago
251 stars

Top 99.8% on SourcePulse

GitHubView on GitHub
Project Summary

This repository offers a curated collection of modular agent skills, designed for integration into AI agent frameworks adhering to the agentskills.io standard. It targets developers and researchers seeking to enhance agent capabilities across diverse domains like frontend development, cloud infrastructure, and AI/VLM tasks. The primary benefit is standardized, reusable skill distribution via the Agent Package Manager (APM), simplifying agent development and enabling robust skill evaluation.

How It Works

The project organizes reusable agent functionalities into discrete, standalone skills, each conforming to the agentskills.io specification. These skills are managed and distributed through a custom Agent Package Manager (APM), allowing for flexible installation globally, per-user, or within specific projects via an apm.yml manifest. A key component is the waxa CLI tool, which provides a rigorous framework for evaluating skill performance and correctness, acting as a critical adoption gate for newly discovered skills.

Quick Start & Requirements

  • Primary install / run command:
    • Install individual skill: apm install -g mizchi/skills/<category>/<skill-name>
    • Add to project: Include in apm.yml under dependencies.apm.
    • Run waxa CLI: npx @mizchi/waxa <eval.yaml>
  • Non-default prerequisites and dependencies:
    • apm (Agent Package Manager)
    • claude CLI on PATH and authenticated (OAuth or ANTHROPIC_API_KEY) for waxa.
    • Deno 2+ required if running waxa from source.
    • Node.js 24+ for the node-sqlite-vec skill.
  • Links: Full waxa reference available at tools/waxa/README.md.

Highlighted Details

  • Comprehensive Frontend Skills: A suite of skills addresses frontend development concerns, including code quality, dependency health, performance, security reviews, and testing posture.
  • AI/VLM Integration: Features skills for image review using vision models and a toolkit (vlmkit) for visual regression, markup synthesis, and design audits.
  • Robust Skill Evaluation: The waxa CLI facilitates iterative skill testing, multi-model comparison, and A/B testing, ensuring quality before adoption.
  • Standardized Distribution: Skills are packaged and distributed via APM, promoting modularity and easy integration into various agent projects.
  • Reproducible Environments: Support for Nix and Devbox via the nix-setup skill aids in creating consistent development environments.

Maintenance & Community

This repository is maintained by @mizchi. No specific community channels (like Discord or Slack), sponsorships, or partnerships are detailed in the provided README.

Licensing & Compatibility

Each skill may have its own LICENSE.txt. Skills lacking an explicit license default to MIT. The MIT license is generally permissive for commercial use and closed-source linking, but users must verify individual skill licenses for compliance.

Limitations & Caveats

License clarity requires checking individual skill LICENSE.txt files, as only unspecified skills default to MIT. Some skills depend on external repositories, which may have separate maintenance statuses or licensing terms. The waxa CLI requires specific setup (Claude CLI, authentication) and version 0.1.0 contained a broken shebang; pinning versions is recommended for reproducibility. Some skills are language-specific (e.g., MoonBit, Gleam) or have localized documentation (SKILL-ja.md).

Health Check
Last Commit

2 weeks ago

Responsiveness

Inactive

Pull Requests (30d)
5
Issues (30d)
2
Star History
66 stars in the last 30 days

Explore Similar Projects

Feedback? Help us improve.