Install agent skills

Root agent skills are authoritative instructions your AI coding agent can load to build on Root correctly. Install once to help prevent your agent — whether that's Claude Code, Cursor, Copilot, or any of ~45 other supported harnesses — from hallucinating Root's conventions, so it can write code that pushes cleanly.

Install

Run this in your project folder. Skills don't auto-update — re-run the command when you want the latest.

npx (universal)

npx skills add https://api.rootplatform.com/v1/insurance/docs --yes

Works with Claude Code, Cursor, Amp, Antigravity, Cline, Codex, Windsurf, GitHub Copilot, Gemini CLI, and every other vercel-labs/skills-compatible agent. Creates a canonical copy under ./.agents/skills/ plus per-agent symlinks (e.g. .claude/skills/…).

Claude Code

Same command as above — Claude Code is a first-class target of the universal installer. No separate flow needed.

Cursor

Same command — Cursor is also a first-class target of the universal installer.

Try it

After installing, restart your agent and try prompts like these. The skill catalogue will route each one through the right reference files.

Build and edit a product module

• "Walk me through creating a new Root product module for pet insurance."
• "Add an alteration hook for adding a spouse to my existing funeral product."
• "How do I migrate my product module to TypeScript?"
• "Write a scheduled function that applies an annual premium increase."

Embed Root into an app

• "Embed a Root quote flow into my React app — what's the right shape?"
• "Author an embed-config.json for a sales flow with custom styling."
• "How do I generate a policy-management portal URL for a specific policyholder?"

Integrate via the Root API

• "Issue a policy end-to-end from my backend using the Root MCP."
• "What's the correct call sequence for quote → application → issue?"
• "I want to list all product modules my organisation has access to."

The agent will load only the reference files it needs per task — reducing manual context-pasting, hallucinated API shapes, and wrong currency units.

What's inside

Three skills are installed. Each is a thin router that knows when to pull in the deep-dive reference files.

• root-platform-best-practices — load-first router. Picks the right deep skill for any Root task, plus cross-cutting basics (MCP install, sandbox vs production, the four-call policy-issuance sequence, the rp / npx @rootplatform/cli command reference).
• root-product-modules — end-to-end lifecycle for authoring product modules: list → clone → edit → test → push → publish. Covers .root-config.json, hooks (quote, application, claim, alteration, lifecycle), schemas, and the TypeScript migration path.
• root-embed-integration — embedding the Root customer flow into a web or mobile app. embed-config.json authoring, the iframe vs standalone vs management-portal-URL decision, compliance and prefill rules.

Verify your install

Inspect the raw catalogue:

curl -s https://api.rootplatform.com/v1/insurance/docs/.well-known/agent-skills/index.json | jq '.skills[] | {name, version, files: (.files | length)}'

You should see three skills with a total of roughly 50 files across them.

How updates flow

The hosted catalogue is served live from Root's documentation source — when that source changes, the catalogue reflects it on the next request (5-minute cache). Your local install doesn't auto-pull; re-run the npx skills add command when you want the latest content.

Each installed SKILL.md also carries a staleness-check instruction. The first time the agent invokes that skill in a conversation, it fetches the live index, compares versions, and tells you if you're behind — a gentle nudge to re-install when a skill's guidance has meaningfully changed.

Install the Root MCP alongside

Skills and the Root MCP server are designed to work together — install both. They solve complementary problems:

• Skills teach your agent how Root works — the workflows, the conventions, the correct sequence of calls, the edge cases.
• The Root MCP server gives your agent the tools it needs to actually act — clone-product-module, create-quote, push-product-module, issue-policy, and so on.

Skills without the MCP means the agent knows what to do but can't do it. The MCP without skills means the agent has tools but doesn't know which to call or in what order. Together, the agent can read, decide, and act — all grounded in Root's own canonical guidance.

{
  "mcpServers": {
    "root-insurance": {
      "command": "npx",
      "args": ["-y", "@rootplatform/mcp-server"],
      "env": {
        "ROOT_INSURANCE_API_KEY": "sandbox_..."
      }
    }
  }
}

See @rootplatform/mcp-server on npm for the full list of tools and configuration options.

Editing prerequisites

The clone-product-module, push-product-module, and publish-product-module tools shell out to the Root workbench CLI (@rootplatform/cli). If you plan to edit modules — not just issue policies — install it once globally:

npm install -g @rootplatform/cli

Most agents (including Claude Code) will prompt to install it the first time they hit a clone/push tool, so this is a convenience step rather than a hard gate.