This lets you codify reasoning strategies that Claude can execute and refine over time. You define patterns like decision trees, chain of thought breakdowns, or analogy mappings, then wrap them into workflows that can be run and evaluated. It's built on the claude-flow MCP server and gives you hooks to track how patterns perform and share the successful ones. Honestly most useful if you're building agent systems where you want consistent, improvable reasoning rather than ad hoc responses. The pattern types are pretty standard cognitive science stuff, but having them as executable primitives with learning feedback is the actual value here.
npx skills add https://github.com/ruvnet/ruflo --skill cognitive-pattern