This is for building always-on wake word detection using openWakeWord, the kind of "Hey JARVIS" activation you'd need for a voice assistant. The skill is built around TDD patterns and privacy constraints: process everything locally, keep CPU under 5%, clear audio buffers immediately after detection, and stay offline. You get concrete tests for things like false positive rates and memory footprint, plus implementation patterns for the actual detector and false positive reduction. It's opinionated about efficiency (targets under 100MB RAM) and assumes you care more about not accidentally recording someone than squeezing out every percentage point of accuracy. If you need cloud-based detection or don't want to think about resource constraints, this probably isn't your approach.
npx skills add https://github.com/martinholovsky/claude-skills-generator --skill wake-word-detection