This one runs worker/verifier cycles until your exit condition hits. You give it a worker prompt that does the work each iteration, then either a shell check for objective pass/fail or a verifier prompt for judgment calls, or both when you want the shell to rule out obvious failures first. It shines when you need something babysat, like keeping a PR green by fixing CI failures every few minutes, or driving tests to pass by iterating on fixes. The cross-provider setup is clever: run your worker on one model and verifier on another so they catch each other's mistakes. Set max iterations and max time unless you enjoy runaway loops.
npx -y skills add getpaseo/paseo --skill paseo-loop --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