Skip to main content
Corbado Observe ships an Agent Skill (the open agentskills.io standard) that teaches your AI coding agent exactly how to instrument Corbado Observe in your frontend: which events to fire, where to fire them, and how to verify the result. The skill uses the SKILL.md format read by most AI coding tools, including Cursor, Claude Code, Codex, GitHub Copilot, Windsurf, and Gemini CLI — plus ~70 more compatible agents. There are two ways to use the skill with your agent, explained below.

​
1. Use it in a prompt

The fastest way — point your agent at the skill directly in a chat prompt, no install required: 
Integrate Corbado Observe into my frontend using https://docs.corbado.com/skill.md

​
2. Install it

To make the skill permanently available to your agent, install it: 
npx skills add https://docs.corbado.com
The https:// scheme is required. The bare domain form (npx skills add docs.corbado.com) is not accepted by the Skills CLI.
The command detects your installed agents and writes the skill into each one’s namespaced skills folder (for example .cursor/skills/, .claude/skills/). It never edits or overwrites your existing AGENTS.md or CLAUDE.md.
Manually installed skills don’t auto-update. Run npx skills update -y to pull the latest version.

​
2.1 Example prompts

Once the skill is installed, ask your agent in natural language:
  • “Integrate Corbado Observe into my frontend”
  • “Add Corbado Observe tracking to my login and signup flows”
  • “Instrument my passkey login with Corbado Observe events”

​
3. What the agent does

The skill guides the agent through a safe, telemetry-only integration:
  1. Discovers the auth surface — finds your login, signup, recovery, and enrollment entry points and the methods each offers.
  2. Proposes a plan — shows you a short map of the touchpoints it will instrument and asks for confirmation before writing code.
  3. Instruments flows, decisions, and subflows with the correct events and payloads, using guarded tracking calls that can never break your app’s behavior.
  4. Verifies by walking each journey with debug mode on and confirming events fire as expected — see the Getting started verify step.

​
4. See also