When your Claude workflow is working but you need it to work faster, cheaper, or more reliably, this is what you reach for. It enforces a proposal-before-building flow where you explore 2-3 optimization directions with tradeoffs, then implement techniques like parallel fan-out, streaming pipelines, model cascading, or complexity-based routing. The mandatory rule is progressive enhancement: every turbocharge must degrade gracefully so the base workflow still functions if the optimization fails. Smart constraint here is refusing to optimize before correctness, which saves you from prematurely optimizing broken logic. Pair it with /evaluate afterward to verify both the performance gain and the graceful degradation.
npx skills add https://github.com/sharpdeveye/maestro --skill turbocharge