boxmoe_header_banner_img

Hello! 欢迎来到阿虪的开发世界~

加载中

文章导读

google/skills:Google 官方 Agent Skills,GCP 与 Gemini 的操作手册一次给齐


avatar
阿虪 8 11 月, 2026 119
google/skills

中文版

google/skills(GitHub:https://github.com/google/skills ,2026-03-31 创建,今晨抓取时已 23,405★,日榜单日涨 443★)是 Google 官方的 Agent Skills 仓库:给 agent 用的 Google 产品与技术操作手册,核心是 Google Cloud。

仓库里有上百个技能,按领域分好类:入门(Google Cloud 上手、IAM 认证授权、基础架构搭建);多产品方案(解决方案架构工作流、多 agent 安全、RAG、跨云智能分析、无边界数据湖仓、在 Google Cloud 上构建与部署 AI agent);AI/ML(Vertex AI Agent Platform 的推理、评估飞轮、RAG 引擎、模型部署与调优、提示词管理、Gemini API、Genkit 各语言版、Live API);基础设施(GKE 集群创建、扩缩容、网络、存储、备份容灾、TPU 排障);数据库与分析(BigQuery、AlloyDB、Spanner、Cloud SQL、数据血缘);开发者工具(gcloud CLI 技能、Cloud Build、Cloud Operator);管理工具(Cloud Monitoring、Cloud Logging、成本分析与优化、SLO 告警);Well-Architected 框架六大支柱(成本、运维卓越、性能、可靠、安全、可持续);安全与身份(SecOps 检测与事件调查、IAM 助手、安全指挥中心);以及 Cloud Run、Firebase、Google Ads API、Google Analytics 等。仓库还汇总了其他 Google 官方技能库(ADK、Android、Dart、Flutter、Firestore、Google Maps)。

安装用 npx skills add google/skills,安装时会让你选技能、目标 agent 与安装范围(项目级或全局),更新用 npx skills update。Claude Code 与 Codex 还能以插件市场的方式装(claude plugin marketplace add google/skills)。Apache-2.0 开源。

为什么值得了解

  • 用 GCP 或 Gemini 的人,这就是官方的操作手册:与其让 agent 自己去翻文档,不如直接给它一套按步骤写的技能,照着做、少踩坑
  • 它和 Anthropic 的官方技能库是同一赛道的两种答案:Google 这一侧重“云上怎么做”,覆盖从开集群到排障、从计费到合规的完整链路;正在求职、做云原生或 AI 应用的人,照着它的分类就能摸清 GCP 的能力地图
  • 对想自己写技能的人也是好教材:每个技能都是一段聚焦的工作流,而不是一堆文档的搬运——好的技能长什么样,照着抄就行

English Version

google/skills (GitHub: https://github.com/google/skills , created 2026-03-31, 23,405 stars at this morning’s fetch, up 443 stars today on the daily trending list) is Google’s official Agent Skills repository: an operations manual for agents covering Google products and technologies, centred on Google Cloud.

It holds over a hundred skills, grouped by area: getting started (onboarding to Google Cloud, IAM authentication and authorisation, foundation building); multi-product solutions (solution-architecture workflow, multi-agent security, RAG, cross-cloud agentic analytics, borderless data lakehouse, building and deploying AI agents on Google Cloud); AI/ML (Vertex AI Agent Platform inference, eval flywheel, RAG engine, model deployment and tuning, prompt management, Gemini API, Genkit in several languages, Live API); infrastructure (GKE cluster creation, scaling, networking, storage, backup and disaster recovery, TPU troubleshooting); databases and analytics (BigQuery, AlloyDB, Spanner, Cloud SQL, data lineage); developer tools (the gcloud CLI skill, Cloud Build, Cloud Operator); management tools (Cloud Monitoring, Cloud Logging, cost analysis and optimisation, SLO alerts); the six pillars of the Well-Architected Framework (cost, operational excellence, performance, reliability, security, sustainability); security and identity (SecOps detection and incident investigation, IAM helpers, Security Command Center); plus Cloud Run, Firebase, the Google Ads API and Google Analytics. The repo also points to other official Google skill libraries (ADK, Android, Dart, Flutter, Firestore, Google Maps).

Install with npx skills add google/skills — it asks which skills, which target agents and the scope (project or global); update with npx skills update. Claude Code and Codex can also install it as a plugin marketplace (claude plugin marketplace add google/skills). Apache-2.0 licensed.

Why it’s worth knowing

  • If you use GCP or Gemini, this is the official operations manual: rather than letting your agent dig through documentation itself, hand it a set of step-by-step skills and it follows them with fewer mistakes
  • It is the counterpart to Anthropic’s official skills library: Google’s side is about “how to do things on the cloud”, covering the full chain from spinning up a cluster to troubleshooting, billing and compliance; for anyone job-hunting or building cloud-native or AI apps, its categories alone map out GCP’s capability landscape
  • It is also good teaching material for writing your own skills: each skill is one focused workflow, not a pile of copied documentation — if you want to know what a good skill looks like, copy this shape


评论(0)

查看评论列表

暂无评论


发表评论

表情 颜文字

插入代码