Agent Skills

Give your coding agent Anlyon's skills: Claude Code, Codex, Cursor, GitHub Copilot and other agents then set up Anlyon, act for your users, define governed actions and handle approvals the way the docs describe. One command to install, MIT licensed.

Your coding agent writes the code that calls Anlyon. Without context it guesses: it invents an endpoint, puts an API key in a source file, or retries a refund whose outcome is unknown. Anlyon’s Agent Skills give it the rules and the exact APIs instead, so the code it writes keeps keys on your server, names the user from your session, and waits for approvals rather than resubmitting.

A skill is a folder with a SKILL.md file and a few reference files, in the open Agent Skills format. Your agent reads each skill’s one-line description at startup and loads the rest when a task matches. It costs almost nothing until it is used.

The skills

Skill	Teaches your agent to
`anlyon-setup`	Add Anlyon to a project: keys and scopes, the SDK or MCP, a test governed call and its receipt
`anlyon-end-users`	Act in each of your users’ own accounts: `forUser`, connect links and the callback, end-user approvals, session tokens, the React components
`anlyon-governed-actions`	Define actions: credentials bound to hosts, impact, read-backs, approval policies, impact limits, OpenAPI import, packs
`anlyon-agent-loop`	Handle results in an agent loop: `pending_approval`, `connect_required`, refusals, `unknown`, idempotency
`anlyon-troubleshooting`	Diagnose a failing call from its error code, and fix it

Each skill does one job, so your agent loads the one it needs and nothing else. Install all five: a skill that does not match the task stays as one line of context.

Install with one command

Works for Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode, Windsurf and other agents that read skills. It needs Node.js 22.20 or later.

npx skills add https://docs.anlyon.com

It lists the five skills, asks which to install and for which agents, then copies or links them into each agent’s skills folder. Pick for yourself with flags:

# Every skill, for Claude Code and Codex, in this project, without prompts
npx skills add https://docs.anlyon.com --skill '*' -a claude-code -a codex -y

# One skill, for every project on this machine
npx skills add https://docs.anlyon.com --skill anlyon-setup -g

# Later: fetch the newest versions
npx skills update

The skills CLI reads the index this site publishes at /.well-known/agent-skills/index.json and checks each download’s SHA-256 digest before it installs it. Commit the installed folder to share the skills with your team.

Claude Code plugin

In Claude Code, the skills also come as one plugin from Anlyon’s marketplace:

/plugin marketplace add https://docs.anlyon.com/skills/marketplace.json
/plugin install anlyon@anlyon

From a shell, the same is claude plugin marketplace add https://docs.anlyon.com/skills/marketplace.json and claude plugin install anlyon@anlyon. Plugin skills are named after the plugin: /anlyon:anlyon-setup. To update, run claude plugin marketplace update anlyon, then claude plugin update anlyon@anlyon.

Install by hand

Download the bundle, or one skill, and unzip it into your agent’s skills folder. Each zip holds one folder per skill.

All five skills

anlyon-skills.zip

One skill at a time

A zip per skill, and the raw files.

curl -fsSLO https://docs.anlyon.com/skills/anlyon-skills.zip
mkdir -p .claude/skills && unzip -o anlyon-skills.zip -d .claude/skills

Invoke-WebRequest https://docs.anlyon.com/skills/anlyon-skills.zip -OutFile anlyon-skills.zip
Expand-Archive anlyon-skills.zip -DestinationPath .claude\skills -Force

Swap .claude/skills for your agent’s folder:

Agent	In one project	For every project
Claude Code	`.claude/skills/`	`~/.claude/skills/`
OpenAI Codex	`.agents/skills/`	`~/.agents/skills/`
Cursor	`.agents/skills/` or `.cursor/skills/`	`~/.agents/skills/` or `~/.cursor/skills/`
GitHub Copilot (VS Code, CLI, cloud agent)	`.github/skills/` or `.agents/skills/`	`~/.copilot/skills/` or `~/.agents/skills/`
Gemini CLI	`.gemini/skills/` or `.agents/skills/`	`~/.gemini/skills/` or `~/.agents/skills/`
OpenCode	`.opencode/skills/` or `.agents/skills/`	`~/.config/opencode/skills/` or `~/.agents/skills/`
Amp, Windsurf, Factory, Goose and other agents that follow the convention	`.agents/skills/`	`~/.agents/skills/`

Each skill must sit in a folder named after it, with SKILL.md directly inside: .claude/skills/anlyon-setup/SKILL.md. Agents change these paths between releases; if a skill does not appear, check your agent’s skills documentation.

Claude apps and the Claude API

Claude and Claude desktop: open Customize → Skills, choose Upload a skill, and upload one skill’s zip at a time. Code execution must be on.
Claude API: upload a skill’s folder with the Skills API, then attach it to the container of a request that uses the code execution tool.

Use them

You do not need to name a skill. Ask for the work, and the agent loads the skill whose description matches:

“Add Anlyon to this Next.js app so the assistant can open issues in each user’s own GitHub, with the user approving each one.”
“Turn our direct Stripe refund call into an Anlyon action with a $500 daily limit.”
“This invoke returns 409 END_USER_NOT_CONNECTED. Fix the chat route.”

In Claude Code you can also call one directly: /anlyon-setup, or /anlyon:anlyon-setup from the plugin.

Skills teach the agent how to write code against Anlyon. To let the agent call your actions itself while you work, also connect it to Anlyon’s MCP server.

Download each skill

Skill	Zip	`SKILL.md`
`anlyon-setup`	anlyon-setup.zip	SKILL.md
`anlyon-end-users`	anlyon-end-users.zip	SKILL.md
`anlyon-governed-actions`	anlyon-governed-actions.zip	SKILL.md
`anlyon-agent-loop`	anlyon-agent-loop.zip	SKILL.md
`anlyon-troubleshooting`	anlyon-troubleshooting.zip	SKILL.md

Each skill’s page lists its reference files.

License

The skills are MIT licensed. Each skill folder carries its LICENSE. Copy them, change them for your codebase, and ship them in your own repository.

What the skills hold to

Keys stay in environment variables. No skill tells an agent to print, log or commit an Anlyon key or a provider credential.
Your server names the user. The model never chooses whose account it acts in.
No blind retries. An unknown outcome goes to a person, and a parked call waits for its decision.
No guessed APIs. Every endpoint, header, scope and error code in the skills is the one in the API reference. Where a skill stops, it points at the page’s Markdown version on this site.