Awesome Skills

A curated library of role-based AI skills, organised by professional domain.

🌐 Website · 📚 Catalog · 📖 Install Guide · 🤝 Contributing

Language: English | 中文

English

What this is

A library of 956 skill files (SKILL.md + optional references/) covering 3 kinds (persona / tool / workflow) across 60+ domains. Each skill is a prompt pack — an instruction set plus reference material — designed to load into Claude Code, OpenCode, Cursor, or similar agent runtimes to give an LLM a consistent professional persona, tool expertise, or engineering workflow methodology.

What this is NOT

• Not an alternative to executable skill packs. The official Anthropic/OpenAI SKILL.md convention can ship with scripts/ and assets/ that the agent runs. The vast majority of skills here are persona + methodology, not runnable tools. Treat them as opinionated system prompts, not as plugins.
• Not a certification body. Historical quality scores in this repo were produced by a self-authored scoring script grading against a self-authored rubric. The scoring is useful internally; it is not an independent benchmark.
• Not feature-complete for every domain. Coverage is broad but uneven — some categories have deep role hierarchies, others have a single stub.

Layout

skills/                   956 skill files organised by kind
  persona/                827 role-based professional persona skills
    <domain>/<role>/
      SKILL.md            Frontmatter + system prompt (target ≤ 300 lines)
      references/         On-demand deep content (workflow / scenarios / …)
      EVALUATION_REPORT.md  Optional self-scored quality report

  tool/                   116 technology-specific expert skills (kind: tool)
    <technology>/
      <skill-name>/SKILL.md

  workflow/               8 process-driven action skills (kind: workflow)
    engineering/          TDD, debug-diagnose, zoom-out, architecture-review,
                          issue-triage, to-prd
    meta/                 write-skill, caveman

external/                 Registry of curated third-party skill repos
                          (anthropics/skills, mattpocock/skills, VoltAgent, …).
                          Pulled on demand via scripts/sync_external.py — not vendored.
benchmarks/               Evaluation dataset + scoring script for comparison
packages/                 Curated skill bundles by domain
roadmap/                  Career-path documents (independent of skills/)
taxonomy.yml              Single source of truth: 21 categories, 80 aliases
tools/                    Python package for skill analysis (scoring, tokens, anti-patterns)
scripts/                  Maintenance scripts (catalog regen, external sync, taxonomy check, …)
.github/                  CI workflows + scripts, CI/CD docs

Quick start

The simplest way to install a skill is to have your agent read its SKILL.md URL:

Read https://github.com/theneoai/awesome-skills/blob/main/skills/persona/executive/ceo/SKILL.md and install as a skill

Platform-specific instructions (OpenCode native command, Claude Code, Cursor, Cline, Codex, Kimi) are in INSTALL-GUIDE.md.

Skill packages

Pre-bundled collections in packages/:

Featured enterprise skills

Role skills modelled after the methodology of specific companies (15 shown out of 100+ under skills/enterprise/):

Full list in CATALOG.md.

External ecosystem hub

external/ is a registry of top third-party skill / subagent / plugin repositories — not a mirror. Pull any subset on demand:

python3 scripts/sync_external.py --list                 # see what's registered
python3 scripts/sync_external.py --all                  # shallow-clone everything
python3 scripts/sync_external.py --slug anthropics-skills

Registered sources (see external/README.md for the full table):

To propose a new source, edit external/sources.yml and open a PR. External clones are gitignored and never linted by our CI — we are a pointer, not an enforcer.

Tooling

The repo ships a Python package for static analysis of skills:

pip install -e ./tools/
python -m tools.skill_analyzer.cli score        # 8-dimension rubric scores
python -m tools.skill_analyzer.cli tokenizer    # Token budget + API cost
python -m tools.skill_analyzer.cli antipattern  # Common mistakes scanner

CI (.github/workflows/quality.yml) runs these on any PR that touches skills/, tools/, or .github/scripts/, and blocks merges whose changed SKILL.md files fall below score/token/description thresholds.

Known limitations

• Text-only skills. Virtually no skill ships executable scripts/ or assets/ — they are persona prompts.
• Description overlap. The role taxonomy spans 60+ domains and 956 skills and many descriptions overlap, which hurts automatic skill-discovery in agent runtimes. A description-similarity linter runs in CI (informational).
• Self-scored quality. EVALUATION_REPORT.md files reflect a self-graded rubric, not external review.
• Three category systems. packages/ (14), roadmap/ (22), and skills/ (60) use different taxonomies — being consolidated.

Documentation

License

MIT — see LICENSE.

中文

项目定位

按专业领域组织的 AI Skill 库，共 956 个技能，按 3 种类型（persona / tool / workflow）覆盖 60+ 领域。每个技能是一份 SKILL.md（+ 可选 references/），本质是角色化的提示词包——一套系统指令加参考资料，用于在 Claude Code / OpenCode / Cursor 等 agent 运行时里给 LLM 加载稳定的专业身份、工具专长或工程工作流方法论。

不是什么

• 不是可执行 Skill 套件。Anthropic/OpenAI 官方 SKILL.md 可以随包携带 scripts/ 和 assets/ 交给 agent 调用；本仓库绝大多数 skill 仅提供角色 + 方法论的文本，应视作有观点的系统提示词，而非可运行插件。
• 不是第三方认证。仓库内历史质量分由本仓库自己编写的脚本按自家标准打出，对内有参考价值，但不是独立基准。
• 不是每个领域都覆盖完整。分类很广但深浅不一：有些领域有完整角色树，有些只有一个占位文件。

目录结构

skills/                   956 个技能文件，按 kind 分层组织
  persona/                827 个角色 persona 技能
    <domain>/<role>/
      SKILL.md            Frontmatter + 系统提示词（目标 ≤ 300 行）
      references/         按需加载的深度内容（workflow / scenarios / …）
      EVALUATION_REPORT.md  可选：自评质量报告

  tool/                   116 个技术工具专家技能（kind: tool）
    <technology>/
      <skill-name>/SKILL.md

  workflow/               8 个流程驱动工作流技能（k