CopilotKit skills and plugin for agents
CopilotKit agent skills
A Claude Code plugin and CopilotKit skills: the agent reads current docs and source, checks a project with verify and sets up Slack channels
- 38,349 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The plugin connects two MCP servers, one of them, nx-mcp, started locally with npx, and the documentation server receives your search queries
- The verify command probes your app's addresses and with the round-trip flag runs the agent, which spends paid model requests
- The Slack setup skill works with Slack app tokens and channel secrets
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/copilotkit-agent-skillsDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The plugin connects two MCP servers, one of them, nx-mcp, started locally with npx, and the documentation server receives your search queries
- The verify command probes your app's addresses and with the round-trip flag runs the agent, which spends paid model requests
- The Slack setup skill works with Slack app tokens and channel secrets
This is third-party code. Review the repository files before installing.
What it does
The CopilotKit repository, a frontend framework for AI agents and generative interfaces, holds a Claude Code plugin and nine skills in the `skills` folder. Five of them are for developers building apps on CopilotKit, and four serve the project's own team. The skills can be installed with `npx copilotkit@latest skills install`. The `copilotkit` plugin is described in `.claude-plugin/plugin.json` and connects the documentation MCP server `https://mcp.copilotkit.ai/mcp`. The `copilotkit` skill deliberately carries no API detail because it changes. Instead it makes the agent read current material first: documentation pages as Markdown (append `.md` to the address), the `llms.txt` index and the source in `node_modules/@copilotkit/`. When the MCP server is loaded, the agent uses the `search-docs`, `search-code`, `search-ag-ui-docs` and `search-ag-ui-code` tools and the browsing `explore-docs` and `explore-code`. The `copilotkit-cli` skill drives the `npx copilotkit@latest` command: `verify` checks the project wiring, `create` scaffolds a new project, and there are sign-in to the Intelligence cloud, onboarding of an existing app and thread-history import. The `verify` command runs up to eleven checks: the project key, the runtime response, the registered agent, thread routes, CORS rules and package version match. Three skills cover Intelligence channels: `channels-setup` leads from nothing to an agent that answers in Slack or Microsoft Teams, `copilotkit-channels` covers the channel code, and `setup-slack-channel` walks through creating a Slack app with tokens and attaching it to a managed channel.
Who it is for. For React and Next.js developers who add a chat and a generative interface to an app with CopilotKit and want the agent to check against the current documentation.
Good fit when
- You want to add a CopilotKit chat, frontend tools or human approval to an app and the agent confuses old and new calls
- A CopilotKit app does not respond and you first need to find the break: key, runtime address, agent, CORS or package versions
- You are connecting an agent on LangGraph, CrewAI, Mastra or PydanticAI and need the right quickstart
- You want to bring an agent into Slack or Microsoft Teams through a managed Intelligence channel
- You need to search the CopilotKit and AG-UI protocol source without cloning the repositories
Not a fit when
- The app does not use CopilotKit: the skills describe only its documentation, command line and channels
- You want an API reference stored inside the skill: it deliberately keeps none and sends the agent to read current pages over the network
- You need a network-isolated environment: the agent reads documentation and source at docs.copilotkit.ai and GitHub, and the `verify` checks call your server
- You need skills for working on the CopilotKit repository itself: four skills about the Inspector panel, Intelligence documentation sections and a terminology list serve work on the project itself
Example request
My CopilotKit chat gets no answer from the LangGraph agent, check the project with verify and tell me where the break isLimitations
- The plugin and skills rely on the current state of the documentation over the network: without access to docs.copilotkit.ai and GitHub the agent has no reference.
- The `.mcp.json` file the plugin points to also connects an `nx-mcp` server with the `npx nx mcp` command. It serves CopilotKit's own monorepo and can be disabled in someone else's project.
- The `verify` command with the `--round-trip` flag runs the agent and spends a model request, so the flag is opt-in. An app without the Intelligence cloud needs `--expect-runtime oss`, otherwise the cloud checks fail.
- The `copilotkit-cli` skill adds a `--via skill:copilotkit-cli` flag to every command, which tells CopilotKit telemetry that the call came from the skill.
- The `setup-slack-channel` skill is built around the OpenTag example and a local run, and Slack and Teams channels work through the Intelligence cloud.
- MIT license.
How to disable. Disable the plugin with `/plugin uninstall copilotkit` in Claude Code and delete the installed skills from the agent's skills folder. The `copilotkit-docs` server goes with the plugin. The `@copilotkit/*` packages in your project are untouched.
FAQ
How do I install the CopilotKit skills?
Run `npx copilotkit@latest skills install` in the project root. The authors suggest doing this once you already work with CopilotKit or have finished setup. Per-agent details are collected on the Build with agents documentation page.
How does the `copilotkit` skill differ from an ordinary API reference?
It carries almost no API detail: the authors hold that a copy in a file starts to drift the day it is written. The skill explains where the current answer lives and demands that the agent look it up in the docs and source before writing code.
What does the verify command check?
Up to eleven things. On the cloud side, a selected Intelligence project and a key that exists and authenticates. In the runtime, the response to a request, a declared agent, use of the key and thread routes, plus acceptance of the browser's origin. The frontend's files and package version match are checked too. The result names the URL probed and where it came from.
Why does verify fail on an app without Intelligence?
By default the command checks cloud features that the open-source setup does not have. For a self-hosted runtime run it with `--expect-runtime oss`, `--round-trip` and `--agent <id>`; `--round-trip` spends a model request. Running `login` or `project select` just to pass the check is not needed.
Which agent frameworks does the skill cover?
The skill names LangGraph, CrewAI, Mastra, ADK and PydanticAI, and the list is not complete. Each framework has its own quickstart in the documentation, and their addresses come from `llms.txt`.
Discussion
No questions or reviews yet.
Related
PluginsAI Builder Club Skills
AI Builder Club — Skills
A Claude Code plugin marketplace to make a repo agent-ready and run compounding agent loops
A CLI skill manager for AI agents: install from repositories, translate between agent formats and deploy into their configs
The author's personal Claude Code toolbox: meta-prompt and todo commands, skills to create skills, hooks and MCP servers, auditors and a Ralph loop
A tool that builds tested skills from a real workflow and governs their lifecycle through your own marketplace
A desktop app and CLI to install, sync and organize agent skills across Claude Code, Codex, Cursor and other tools
A set of skills, slash commands and agents for Claude Code that helps you build your own skills, agents, hooks, slash commands and prompts
README in short
The CopilotKit description opens with a prompt for an agent: an onboarding command run from the project root. It continues with creating new projects, a skills section, the framework's features (chat, generative interface, shared state, human approval, automatic learning) and documentation links for each. The repository holds a plugin manifest, a Claude Code marketplace file, nine skills with references and example apps, some with their own skills under `examples`.
README badge
Original SKILL.md
--- name: copilotkit description: "Use for any CopilotKit question — adding it to an app, chat UI, frontend or server tools, generative UI, shared state, human-in-the-loop, agent frameworks (LangGraph, CrewAI, Mastra, ADK, PydanticAI, and others), the runtime, Intelligence, threads, voice, or diagnosing something that is not working. Do not answer from memory: this skill exists to point you at the current documentation and source, both of which you can read." version: 3.2.0 --- # CopilotKit CopilotKit's APIs move. Anything written down in a skill file is a copy that starts drifting the day it is written, so this skill carries almost no API detail on purpose. It tells you where the current answer lives and how to get it. **Look it up before you write code.** Not because the docs are more convenient, but because they are regenerated from the source and a recollection is not.