Awesome Skills
A curated list of Agent Skills, resources, and tools for AI coding agents like Claude Code, Codex, Gemini CLI, GitHub Copilot, and more.
Ask AI about Awesome Skills
Powered by Claude · Grounded in docs
I know everything about Awesome Skills. Ask me about installation, configuration, usage, or troubleshooting.
0/500
Reviews
Documentation
Awesome Skills
A curated library of role-based AI skills, organised by professional domain.
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.mdconvention can ship withscripts/andassets/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/:
| Package | Focus |
|---|---|
| tech | Software, AI/ML, data |
| executive | CEO, CTO, CFO, COO, CMO |
| finance | Banking, consulting, investment |
| healthcare | Clinical and medical management |
| software | Backend, frontend, devops, QA |
Featured enterprise skills
Role skills modelled after the methodology of specific companies (15 shown out of 100+ under skills/enterprise/):
| Skill | Company | Methodology |
|---|---|---|
| amazon-engineer | Amazon | 14 LPs, Working Backwards, 6-page memos |
| tesla-engineer | Tesla | First principles, five-step algorithm |
| spacex-engineer | SpaceX | Rapid iteration, cost innovation |
| nvidia-ml-engineer | NVIDIA | CUDA optimisation, GPU platforms |
| mckinsey-consultant | McKinsey | MECE, issue trees, pyramid principle |
| toyota-engineer | Toyota | TPS, JIT, Kaizen, Jidoka |
| anthropic-researcher | Anthropic | Constitutional AI, interpretability |
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):
| Category | Repos |
|---|---|
| Official (Anthropic) | anthropics/skills, claude-plugins-official, knowledge-work-plugins |
| Curated lists | VoltAgent/awesome-agent-skills, hesreallyhim/awesome-claude-code, travisvn/awesome-claude-skills, ComposioHQ/awesome-claude-skills |
| Subagents & orchestration | VoltAgent/awesome-claude-code-subagents, wshobson/agents, 0xfurai/claude-code-subagents |
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/orassets/— 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.mdfiles reflect a self-graded rubric, not external review. - Three category systems.
packages/(14),roadmap/(22), andskills/(60) use different taxonomies — being consolidated.
Documentation
| Doc | Purpose |
|---|---|
| CATALOG.md | Full catalog of all skills |
| INSTALL-GUIDE.md | Platform-specific install instructions |
| CONTRIBUTING.md | How to add or improve skills |
| .github/CI.md | CI/CD pipeline notes |
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（kind: workflow）
engineering/ TDD、debug-diagnose、zoom-out、architecture-review、
issue-triage、to-prd
meta/ write-skill、caveman
external/ 精选第三方 skill 仓库的注册表
（anthropics/skills、mattpocock/skills、VoltAgent 等）
通过 scripts/sync_external.py 按需拉取，不随仓库 vendored
benchmarks/ 评估数据集 + 评分脚本
packages/ 按领域打包的 skill 合集
roadmap/ 职业路径文档（独立于 skills/）
taxonomy.yml 21 个分类、80 个别名的唯一真源
tools/ Skill 分析工具 Python 包（评分、token、反模式）
scripts/ 维护脚本（catalog 重生成、external 同步、taxonomy 检查等）
.github/ CI 工作流与脚本、CI/CD 文档
快速开始
最通用的安装方式是让 agent 读 SKILL.md 的 URL：
Read https://github.com/theneoai/awesome-skills/blob/main/skills/persona/executive/ceo/SKILL.md 并安装为 skill
各平台（OpenCode 原生命令、Claude Code、Cursor、Cline、Codex、Kimi）详细步骤见 INSTALL-GUIDE.md。
Skill 合集
按领域打包，见 packages/：
| 合集 | 覆盖 |
|---|---|
| tech | 软件、AI/ML、数据 |
| executive | CEO、CTO、CFO、COO、CMO |
| finance | 银行、咨询、投资 |
| healthcare | 临床与医疗管理 |
| software | 后端、前端、devops、QA |
精选企业技能
仿照特定公司方法论建模的角色（从 skills/enterprise/ 100+ 条中选 7 条）：
| Skill | 公司 | 方法论 |
|---|---|---|
| amazon-engineer | Amazon | 14 条领导力准则、Working Backwards、6 页备忘录 |
| tesla-engineer | Tesla | 第一性原理、五步算法 |
| spacex-engineer | SpaceX | 快速迭代、成本创新 |
| nvidia-ml-engineer | NVIDIA | CUDA 优化、GPU 平台 |
| mckinsey-consultant | McKinsey | MECE、Issue Tree、金字塔原理 |
| toyota-engineer | Toyota | TPS、JIT、改善、自働化 |
| anthropic-researcher | Anthropic | Constitutional AI、可解释性 |
完整列表见 CATALOG.md。
外部生态集散地
external/ 是业界优秀 skill / subagent / plugin 仓库的注册表（非镜像），按需拉取任意子集：
python3 scripts/sync_external.py --list # 查看注册的仓库
python3 scripts/sync_external.py --all # 全量浅拉取
python3 scripts/sync_external.py --slug anthropics-skills
已注册仓库（完整表格见 external/README.md）：
提议新增来源：编辑 external/sources.yml 并提交 PR。
external/ 下的本地克隆已加入 .gitignore，也不被 CI 扫描——我们是指针，不是上游的审阅者。
工具
仓库附带 Python 分析包：
pip install -e ./tools/
python -m tools.skill_analyzer.cli score # 8 维度评分
python -m tools.skill_analyzer.cli tokenizer # Token 预算与 API 成本
python -m tools.skill_analyzer.cli antipattern # 反模式扫描
CI（.github/workflows/quality.yml）会在修改了 skills/、tools/ 或 .github/scripts/ 的 PR 上运行这些工具，若新改动的 SKILL.md 达不到阈值则阻塞合并。
已知局限
- 绝大多数 skill 是纯文本 persona，并不随包携带可执行
scripts/或assets/。 - 描述重叠严重：60+ 分类 × 956 个技能，许多 description 字段相似度高，会降低 agent 运行时的 skill 自动发现准确度。描述相似度检查已在 CI 中运行（仅提示，不阻塞）。
- 质量分是自评：
EVALUATION_REPORT.md反映仓库自家规则的评分，不是独立评审。 - 三套分类互不一致：
packages/（14 类）、roadmap/（22 类）、skills/（60 类）仍在统一中。
文档
| 文档 | 用途 |
|---|---|
| CATALOG.md | 完整技能目录 |
| INSTALL-GUIDE.md | 各平台安装步骤 |
| CONTRIBUTING.md | 如何贡献或改进技能 |
| .github/CI.md | CI/CD 说明 |
相关项目
- Awesome MCPs - 115+ MCP 服务器，一键安装
- GitHub - Awesome MCPs 项目源码
许可证
MIT，见 LICENSE。