After you've used an AI agent skill and learned what works or what's missing, this lets you fold that feedback back into the skill's source. It handles the full loop: listen to your notes, propose concrete edits to the SKILL.md, apply them, commit, and re-install via agr upgrade. For in-repo skills it edits directly. For remote ones it offers to fork locally or file a GitHub issue upstream. It's explicit about not interrupting you with checklists, instead taking whatever shape of feedback you offer, whether that's a single sentence or a full retrospective. The boundaries are clear: greenfield authoring and skill installation live elsewhere.
npx -y skills add computerlovetech/agr --skill skill-debrief --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