For Users & Teams

Install verified skills

One command installs domain-verified skills into any AI tool. No pre-install required. No guessing who published it.

// install in one command

npx agent-root install

No global install. No manual mkdir + curl. Just run one command — it finds the skill, downloads the SKILL.md, and puts it in the right directory for your tool.

# Install a skill (auto-detects your AI tool) npx agent-root install stripe.com/payments # Specify a tool explicitly npx agent-root install stripe.com/payments --tool claude # Project-scoped install (shared with your team via git) npx agent-root install stripe.com/payments --project # Install all skills from a domain npx agent-root install stripe.com --all

Search the registry from your terminal

npx agent-root search "github actions"

No Node.js? Use curl

curl -sf https://agentroot.io/install.sh | sh -s -- stripe.com/payments # With options: curl -sf https://agentroot.io/install.sh | sh -s -- stripe.com/payments --tool claude --project

Manage installed skills

# List all installed AgentRoot skills npx agent-root list # Update a skill (re-fetches SKILL.md from source) npx agent-root update stripe.com/payments # Remove a skill npx agent-root uninstall payments
How it works: When --tool is omitted, agentroot auto-detects which AI tools you have installed by checking for ~/.claude/, ~/.codex/, ~/.gemini/, and .cursor/. If none are found, it falls back to .agents/skills/ — the cross-tool standard.

Already published skills?

If you've registered your domain on AgentRoot with _skill DNS records, your skills are already installable via npx agent-root install yourdomain.com/your-skill. No extra work needed — this page is for the consumer side.

// find a skill

Search the registry

$ find
Browse all →
// per-tool guides

Install in your AI tool

Claude Code

Claude Code reads SKILL.md files from personal and project skill directories. The CLI handles both.

Install

# Personal skill (available in all your projects) npx agent-root install example.com/github-search --tool claude # Project skill (committed to repo, shared with team) npx agent-root install example.com/github-search --tool claude --project

Use it

The skill is immediately available. Claude reads the skill's description and auto-invokes when relevant.

# Just ask naturally — Claude finds the right skill Find me trending AI repos on GitHub this week

Skill locations

ScopePathApplies to
Personal~/.claude/skills/<name>/SKILL.mdAll your projects
Project.claude/skills/<name>/SKILL.mdThis repo only
Tip: Claude reads the description field in SKILL.md frontmatter to decide when to auto-invoke a skill. Well-written descriptions mean skills "just work."

OpenAI Codex CLI

Codex CLI reads SKILL.md from ~/.codex/skills/ (global) or .agents/skills/ (project-level).

Install

# Global skill npx agent-root install example.com/weather-api --tool codex # Project-level skill (uses .agents/skills/ cross-tool standard) npx agent-root install example.com/weather-api --tool codex --project

Use it

Restart Codex CLI. The skill is discovered automatically.

codex "What's the weather in Tokyo?"

Skill locations

ScopePathApplies to
Global~/.codex/skills/<name>/SKILL.mdAll projects
Project.agents/skills/<name>/SKILL.mdThis repo only
Note: Codex CLI requires a restart after adding new skills. The .agents/skills/ directory is the cross-tool standard — it works in Codex, Gemini CLI, and others.

Google Gemini CLI

Gemini CLI supports both .gemini/skills/ and the cross-tool .agents/skills/ directories.

Install

# Installs to .agents/skills/ (cross-tool standard) npx agent-root install example.com/code-review --tool gemini # Project-scoped npx agent-root install example.com/code-review --tool gemini --project

Use it

Gemini scans skills at session start and auto-activates them when your request matches.

gemini "Review the changes in my last commit"

Skill locations

ScopePathApplies to
Project (cross-tool).agents/skills/<name>/SKILL.mdThis repo (Gemini + Codex)
Project (Gemini).gemini/skills/<name>/SKILL.mdThis repo (Gemini only)
Tip: Use the .agents/skills/ path if your team uses multiple AI tools. It's the cross-tool standard recognized by Codex CLI and Gemini CLI.

Cursor / Windsurf

IDE-based AI assistants use rules files rather than the SKILL.md directory convention. The CLI downloads to .cursor/skills/ — you can reference the file from your rules, or paste the content directly.

Install

# Downloads SKILL.md to .cursor/skills/deploy/ npx agent-root install example.com/deploy --tool cursor

Use it

Reference the downloaded skill from your .cursorrules or .cursor/rules file, or paste the SKILL.md content directly into your rules.

Note: Cursor and Windsurf don't auto-discover SKILL.md files from directories natively (yet). The CLI downloads the file for you — you still need to reference it from your rules file.

Raw API — build your own

Build custom integrations using the AgentRoot API. The CLI uses these same endpoints under the hood.

Search for skills

# Via CLI (pretty-printed with install commands) npx agent-root search "payments" --json # Via API directly curl "https://agentroot.io/api/find-skills?q=payments" # Response { "skills": [ { "skill_id": "stripe-payments", "name": "Stripe Payments", "description": "Process payments, create invoices, manage subscriptions", "skill_md_url": "https://example.com/.agents/skills/stripe-payments/SKILL.md", "domain": "example.com" } ], "count": 1 }

Fetch the SKILL.md

# Download the SKILL.md for a specific skill curl "https://example.com/.agents/skills/stripe-payments/SKILL.md"

Combined discovery (agents + skills)

# Search both agents and skills in one call curl "https://agentroot.io/api/discover?q=finance" # Get all skills for a specific domain curl "https://agentroot.io/api/skills/example.com"
Full API reference: See api.html for all endpoints, parameters, and response formats.
// can't find what you need?

Publish your own skills

Have an API or tool that other agents could use? Publish it as a SKILL.md and make it discoverable to every AI tool on the internet. DNS-based — no accounts, no approval, no vendor lock-in.

Publish your skills → API docs →