This lets Claude check learned behaviors stored in your project's instinct files before starting work. Run it when you're about to code, test, or commit to surface relevant patterns you've captured, like "always run linters before push" or "prefer integration tests for API routes". Each instinct has a trigger, confidence level, and domain tag so you can quickly filter what matters. The confidence scores (0.3 to 0.9+) tell you how strongly to weight each one. It's basically a way to build up project-specific muscle memory without hardcoding rules everywhere. The bash loops read from `.claude/homunculus/instincts/` and dump everything matching your current context.
npx skills add https://github.com/humanplane/homunculus --skill instinct-apply