Agent Graphs

This walks you through building multi-agent workflows in LaunchDarkly by creating directed graphs where each node is a config (an agent with its own model and prompt) and edges define how they route to each other. You design the topology, set a root config as entry point, and define handoff logic that passes context between agents. Good for triage systems that route to specialists, escalation chains, or any pipeline where different agents handle different steps. The skill is methodical about making you map things out before building, which is smart because these graphs can get messy fast. Requires the LaunchDarkly MCP server and expects you to create the individual agent configs first before wiring them together.

Install to Claude Code

npx -y skills add launchdarkly/agent-skills --skill agent-graphs --agent claude-code