GrowthBook

A prompt is a suggestion. A skill is a procedure. When you type "create a feature flag" into Claude or Cursor, the agent can already call the GrowthBook API, but it doesn’t know your own procedures. New flags ship enabled at 100%. "Analyze this experiment" returns a lift number without checking for a sample ratio mismatch. The output looks good most of the time, so you move on and only find out what the agent skipped later. Our new skills are open-source markdown playbooks. You can read them, fork them, and contribute to them. With these skills, new flags ship disabled by default, changes move through draft → review → publish, and analysis runs an SRM check before reporting any lift. It's not a guardrail you hope your agent remembers. It's a written step in a file it references every time. They run in Claude, Cursor, Codex, or any platform that uses the Agent Skills standard. 👉️ Install them with: npx skills add growthbook/skills