PlugWiki

Claude skills for AI agents

Claude skills for building, testing and deploying AI agents, and for running several agents at once.

Picks
10
ranked

The most installed Claude skills for AI agents teach Claude to build and ship agents on one cloud: Microsoft Foundry on Azure and Google's agents-cli suite. For running many agents at once, Superpowers' subagent-driven-development and dispatching-parallel-agents lead.

Each cloud skill works with that vendor's tools and an account, and deploying creates real resources that can cost money. If you meant "agent skills" as the skills format itself, start at the skills directory, or use skill-creator to write your own.

Good for

  • Developers building agents on Azure, Google Cloud, Cloudflare or LangChain
  • Engineers who want to run several coding agents in parallel
  • Teams that need to test and evaluate an agent before shipping
  • Tinkerers who want Claude to coordinate subagents on a plan

Top picks

Skills about building, testing, deploying or running AI agents, ordered by installs on skills.sh as of November 2, 2026, with GitHub stars beside them. Stars belong to the whole repo, so skills from one repo share a count.

  1. 1
    microsoft-foundrySkill

    Guides Claude through building, deploying, testing and evaluating agents on Microsoft Foundry. Needs an Azure account and a signed-in Azure CLI. Expect real resources and cost.

    1,609 GitHub stars
  2. 2
    google-agents-cli-adk-codeSkill

    Code reference for writing agents with Google's Agent Development Kit in Python or Go. Part of the agents-cli suite, which needs agents-cli installed.

    6,276 GitHub stars
  3. 3
    google-agents-cli-deploySkill

    Deploys agents to Google Cloud: Agent Runtime, Cloud Run or GKE. Its rule: never deploy without explicit human approval, because deploys can cost money.

    6,276 GitHub stars
  4. 4
    google-agents-cli-evalSkill

    Tests an agent's behavior with a loop: prepare data, run the eval, analyze failures, fix and re-run. Starts with one or two cases.

    6,276 GitHub stars
  5. 5
    subagent-driven-developmentSkill

    Runs a written plan by giving each task to a fresh subagent, then reviews every task and the whole branch. Part of Superpowers, so install the whole set.

    310,635 GitHub stars
  6. 6
    dispatching-parallel-agentsSkill

    Splits two or more independent problems across parallel subagents, then reviews and merges the results. Not for related failures or tasks that edit the same files.

    310,635 GitHub stars
  7. 7
    agents-sdkSkill

    Cloudflare's skill for building and debugging apps on the Cloudflare Agents SDK. Tells the assistant to look things up first instead of trusting old training data.

    3,244 GitHub stars
  8. 8
    herdrSkill

    Lets Claude open panes and hand work to other agents inside Herdr, a terminal multiplexer for coding agents. Needs Herdr installed and Claude running in a Herdr pane.

    47,289 GitHub stars
  9. 9
    deep-agents-coreSkill

    LangChain guide for building Deep Agents in Python or TypeScript: agent setup, middleware, skills and storage. Needs the deepagents package and a model provider key.

    1,320 GitHub stars
  10. 10
    rufloSkill

    Runs teams of Claude Code agents with shared memory. The skill is only a pointer; the tools come from the ruflo package, and setup adds hooks. Its install count is low next to its 80,533 stars.

    80,533 GitHub stars

Build, test and ship a first agent with Claude

Claude

  1. 1

    Match the skill to your stack

    Use the Google agents-cli skills for Google Cloud, microsoft-foundry for Azure, agents-sdk for Cloudflare and deep-agents-core for LangChain. Each covers one stack, so pick the one you already have an account for.

    Uses microsoft-foundry, google-agents-cli-adk-code, agents-sdk, deep-agents-core

  2. 2

    Install the suite

    npx skills add https://github.com/google/agents-cli --skill google-agents-cli-adk-code

    The Google skills work together and need the agents-cli tool, so install the other skills in the suite too. Each skill's page lists the exact steps.

    Uses google-agents-cli-adk-code

  3. 3

    Build and test

    google-agents-cli-adk-code says not to write agent code until a project is scaffolded. google-agents-cli-eval says to start with one or two test cases and expect many iterations.

    Prompt
    Create an agent that answers questions about our refund policy. Then write an eval with two test cases and run it.

    Tip: Example prompts are illustrations; swap in your own agent.

    Uses google-agents-cli-eval

  4. 4

    Ask before you deploy

    google-agents-cli-deploy compares Agent Runtime, Cloud Run and GKE. Its rule is to never deploy without explicit human approval, since deploys create real cloud resources.

    Prompt
    Compare the deploy options for this agent and what each one costs. Do not deploy yet.

    Uses google-agents-cli-deploy

Comparison

Our advice: match the skill to the cloud you already use, since each cloud skill assumes that vendor's account and tools. For running a plan across several agents on any stack, start with subagent-driven-development.

SkillStackNeedsInstalls
microsoft-foundryAzureAzure account585,685
agents-cli (Google)Google Cloudagents-cli533,797
subagent-driven-developmentThis pageAnySubagent support244,830
agents-sdkCloudflareagents package112,293
deep-agents-coreLangChainProvider key16,280

Questions

What are Claude agent skills?

A skill is a SKILL.md file of instructions that your assistant loads when a task matches. This list covers skills about AI agents. For the whole skills catalog, see the skills directory.

What is the best Claude skill for building AI agents?

By installs on skills.sh, Microsoft's microsoft-foundry leads, then Google's agents-cli skills. The best one is the skill for the cloud or framework you already use.

How do I write my own agent skill?

Use skill-creator. Claude interviews you about the task, writes the SKILL.md and tests it against a run without the skill. It works fully in Claude Code and partly in Claude.ai. The guide to creating Claude skills covers writing one by hand.

How do I find more skills?

Install find-skills. When you ask how to do something, it searches the skills.sh leaderboard with the skills CLI and shows matches with install counts. It installs something only after you say yes.

Can Claude run several agents at once?

Yes. subagent-driven-development gives each task to a fresh subagent and reviews the result. dispatching-parallel-agents runs independent problems side by side, and says not to use it when failures are related or agents would edit the same files.

Do these skills need a cloud account?

The cloud skills do. microsoft-foundry needs an Azure account and a signed-in Azure CLI, and the Google deploy skill needs a Google Cloud project. deep-agents-core needs a model provider key for your own project.

Can an agent skill deploy things without asking?

The Google deploy skill says never to deploy without explicit human approval, because deploys create real resources and can cost money. microsoft-foundry works on real Azure resources too, so review each step.

Do these skills work in Codex or Cursor?

Several do. The Google agents-cli skills list Antigravity CLI, Claude Code and Codex. Herdr lists Claude Code, with a workaround for Codex, and Ruflo lists Claude Code, Codex, Cursor, Windsurf and others.

Are agent skills free?

The skills are free and open licensed, mostly MIT or Apache-2.0. The clouds and model providers they use are not, so check pricing before you deploy.

Sources

Docs, issue trackers and discussions this page is built from.

Show all 22
  1. skills.sh · skills.sh
  2. skills.sh: microsoft-foundry · skills.sh
  3. microsoft/azure-skills · github.com
  4. skills.sh: google-agents-cli-adk-code · skills.sh
  5. google/agents-cli · github.com
  6. skills.sh: subagent-driven-development · skills.sh
  7. obra/superpowers · github.com
  8. skills.sh: agents-sdk · skills.sh
  9. cloudflare/skills · github.com
  10. skills.sh: google-agents-cli-deploy · skills.sh
  11. skills.sh: google-agents-cli-eval · skills.sh
  12. skills.sh: dispatching-parallel-agents · skills.sh
  13. skills.sh: herdr · skills.sh
  14. herdrdev/herdr · github.com
  15. skills.sh: deep-agents-core · skills.sh
  16. langchain-ai/langchain-skills · github.com
  17. skills.sh: ruflo · skills.sh
  18. ruvnet/ruflo · github.com
  19. skills.sh: skill-creator · skills.sh
  20. anthropics/skills · github.com
  21. skills.sh: find-skills · skills.sh
  22. vercel-labs/skills · github.com

Updated Nov 3, 2026.