Skip to content

Skill Collections

Open-source skill collections and managers for AI agents. The underlying format is the open Agent Skills standard, shared by Claude Code, Codex and other agents. Compatibility depends on each collection and the target agent; installation examples below name the tool they apply to.

  • Superpowers — kompletní agentic workflow framework od Jesseho Vincenta (Prime Radiant); kompozitní “skills” řídí celý vývojový cyklus: brainstorming → plán → implementace přes subagenty → TDD → code review → merge/PR. Agent automaticky detekuje relevantní skill a dodržuje ho jako povinný workflow. Funguje s Claude Code, Cursor, Codex, OpenCode, Gemini CLI a GitHub Copilot CLI.

    Install in Claude Code:

    Terminal window
    /plugin install superpowers@claude-plugins-official

    Klíčové skills: brainstorming, writing-plans, subagent-driven-development, test-driven-development, systematic-debugging, using-git-worktrees, requesting-code-review, writing-skills

  • agent-skills — 19 production-grade engineering workflow skills for AI coding agents covering the full dev lifecycle (define → plan → build → verify → review → ship); works with Claude Code, Cursor, Gemini CLI, Windsurf, and GitHub Copilot

    Install in Claude Code:

    Terminal window
    /plugin marketplace add addyosmani/agent-skills
    /plugin install agent-skills@addy-agent-skills

    Key skills: spec-driven-development, test-driven-development, security-and-hardening, performance-optimization, code-review-and-quality, shipping-and-launch

  • mattpocock/skills — pragmatic engineering skills by Matt Pocock (Total TypeScript / AI Hero); addresses four common agent failure modes — misalignment (/grill-me, /grill-with-docs for upfront requirements interrogation), verbosity (shared-language CONTEXT.md + ADRs built from grilling), broken feedback loops (/tdd red-green-refactor, /diagnose debugging), and architectural drift; small composable skills, model-agnostic; includes a /setup-matt-pocock-skills wizard that asks for issue-tracker preference (GitHub/Linear/local), triage labels, and doc location

    Install:

    Terminal window
    npx skills@latest add mattpocock/skills
  • caveman — “why use many token when few do trick” — skill/plugin that makes the agent talk like a caveman, cutting ~75% of output tokens while keeping full technical accuracy; includes intensity levels (Lite / Full / Ultra), 文言文 classical Chinese mode, terse commit messages (caveman-commit), one-line code reviews (caveman-review), and a caveman-compress tool that rewrites memory files (e.g. CLAUDE.md) into compressed prose saving ~45% of input tokens

    Install in Claude Code:

    Terminal window
    claude plugin marketplace add JuliusBrussee/caveman
    claude plugin install caveman@caveman

    Any agent (Cursor, Copilot, Windsurf, Cline, Codex):

    Terminal window
    npx skills add JuliusBrussee/caveman
  • claude-improve — self-improving retrospective skill; after each conversation Claude reflects on what went wrong and updates its own CLAUDE.md rules to do better next time

  • Chops — macOS desktop app to browse, edit, and organize agent skills across Claude Code, Cursor, Codex, Windsurf, and Amp from a single interface; open source (FSL-1.1-MIT)

Reusable CLAUDE.md behavioral guidelines that can be merged into any project’s configuration.

  • andrej-karpathy-skills — single CLAUDE.md derived from Andrej Karpathy’s observations on common LLM coding pitfalls; enforces four principles: think before coding (surface assumptions and tradeoffs), simplicity first (no speculative abstractions), surgical changes (touch only what’s needed), and goal-driven execution (define verifiable success criteria before implementing)
  • Graphify — knowledge graph builder skill (Tree-sitter + LLM extraction) for codebase understanding
  • Claude Code — commands and settings reference
  • Plugins — Claude Code plugin marketplace