# agent-skills — Production Skills for AI Coding Agents

> Production-grade engineering skills for AI coding agents, with install recipes for Gemini CLI and guidance for Cursor/Codex-style workflows.

## Install

Save the content below to `.claude/skills/` or append to your `CLAUDE.md`:

## Quick Use

1. Install for Gemini CLI (from repo):
   ```bash
   gemini skills install https://github.com/addyosmani/agent-skills.git --path skills
   ```
2. Or install from a local clone:
   ```bash
   git clone https://github.com/addyosmani/agent-skills.git
   gemini skills install ./agent-skills/skills/
   ```

## Intro

Production-grade engineering skills for AI coding agents, with install recipes for Gemini CLI and guidance for Cursor/Codex-style workflows.

- **Best for:** teams who want consistent, reusable engineering “moves” for agents: planning, debugging, refactors, reviews, and guardrails
- **Works with:** Gemini CLI (native install); other agents that can load instruction files (Cursor/Codex/etc.)
- **Setup time:** 15–35 minutes

## Practical Notes

- Quant: start by installing 5–8 skills that cover your highest-frequency tasks (planning, tests, reviews) before importing the full library.
- Quant: track “skill hit rate”: out of 20 tasks, how many used at least one skill file end-to-end?

---

## Make skills composable

Skills work best when they’re small and opinionated:

- One skill = one job (e.g., “plan tasks”, “add tests”, “review PR”).
- Keep inputs explicit (file paths, commands, constraints).
- Make outputs auditable (checklists, diffs, artifacts).

## How to roll out without chaos

1. Choose a small set of baseline skills everyone uses.
2. Put them under version control and review changes.
3. Add new skills only when a real workflow needs it.

## Avoiding prompt drift

Pin your skill versions in a repo and document the expected behavior. If outcomes change, treat it like a dependency upgrade: evaluate, then roll forward.

### FAQ

**Q: Do I need Gemini CLI to benefit?**
A: No. Gemini CLI is one supported install path; any agent that can read instruction files can use the skills.

**Q: Should I install everything?**
A: Start with a small baseline set so behavior is predictable.

**Q: How do I measure impact?**
A: Measure time-to-plan and defect rate before/after introducing skills on the same task type.

## Source & Thanks

> Source: https://github.com/addyosmani/agent-skills
> License: MIT
> GitHub stars: 40,009 · forks: 4,406

---

<!-- ZH -->

## 快速使用

1. 给 Gemini CLI 安装（直接从仓库）：
   ```bash
   gemini skills install https://github.com/addyosmani/agent-skills.git --path skills
   ```
2. 或从本地克隆安装：
   ```bash
   git clone https://github.com/addyosmani/agent-skills.git
   gemini skills install ./agent-skills/skills/
   ```

## 简介

面向 AI 编码 Agent 的工程化技能库：提供可直接安装的技能文件、流程化最佳实践与落地示例，并给出 Gemini CLI 等环境的安装方式，方便在 Cursor/Codex 等工具链中统一执行标准。

- **适合谁：** 希望把规划/调试/重构/评审等工程动作标准化，并复用给多个 Agent 的团队
- **可搭配：** Gemini CLI（原生安装）；以及能加载指令文件的其他 agent（Cursor/Codex 等）
- **准备时间：** 15–35 分钟

## 实战建议

- 量化建议：先挑 5–8 个最高频任务相关的 skills（规划/测试/评审等）安装落地，再逐步扩展到全量。
- 量化建议：统计“技能命中率”：在 20 个任务中，有多少次完整使用了至少 1 个 skill 文件？

## 让 skills 可组合、可复用

最有效的 skills 往往是“小而明确”的：

- 一个 skill 只做一件事（规划、加测试、评审等）。
- 输入要显式（文件路径、命令、约束）。
- 输出要可审计（清单、diff、产物）。

## 如何团队化落地而不失控

1. 先选一小组所有人都用的 baseline skills。
2. 纳入版本管理并走 review。
3. 只有真实工作流需要时再引入新 skills。

## 避免提示词漂移

把 skill 版本 pin 在仓库中并记录预期行为；当结果变化时按依赖升级处理：评估 → 再推进。

### FAQ

**必须用 Gemini CLI 才能用吗？**
答：不需要。Gemini CLI 只是其中一种安装方式；能读取指令文件的 agent 都能用。

**要一次装全吗？**
答：不建议。先落地一小组 baseline skills，效果更可控。

**怎么衡量收益？**
答：在同类任务上对比引入 skills 前后的规划耗时与缺陷率。

## 来源与感谢

> Source: https://github.com/addyosmani/agent-skills
> License: MIT
> GitHub stars: 40,009 · forks: 4,406


---
Source: https://tokrepo.com/en/workflows/agent-skills-production-skills-for-ai-coding-agents
Author: Skill Factory