This overrides Claude's default helpful-but-surface-level mode and makes it actually think strategically before answering. It triggers on phrases like "what should we do" or "I'm stuck" and forces a deeper process: reframe the real problem, map concepts across multiple domains, find where they intersect for non-obvious leverage, and stress-test before recommending. The core move is 4D chess thinking, where you're looking for the single change that solves now and unlocks future capabilities. It's designed to resist the first cached answer and push toward elegant simplicity, the kind that makes complexity dissolve rather than just managing it. Use this when you want Claude to suggest strategy, not just execute tasks.
npx -y skills add pproenca/dot-skills --skill think --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
juliusbrussee/caveman
mattpocock/skills
shadcn/improve
obra/superpowers
forrestchang/andrej-karpathy-skills
vercel-labs/skills