This walks you through creating AI Configs in LaunchDarkly, handling the choice between agent mode (for frameworks like LangGraph and CrewAI that expect instruction strings) and completion mode (for direct API calls with message arrays). It uses the LaunchDarkly MCP server to set up configs with model selection, prompts, and parameters in one shot. The workflow is opinionated about moving fast with sensible defaults rather than asking for every detail. One critical gotcha: newly created configs default to a disabled variation in their fallthrough, so you need to explicitly enable targeting before the SDK will serve your variation. Without that step, every evaluation returns enabled equals false and looks broken when it isn't.
npx skills add https://github.com/launchdarkly/agent-skills --skill aiconfig-create