This is a two-stage learning system that watches for correction patterns during your coding sessions and queues them up, then lets you manually review and apply them to CLAUDE.md files with the /reflect command. It catches things like "no, use X not Y" or tool rejections and stores them in ~/.claude/learnings-queue.json until you're ready to process them. The manual review step is smart because you stay in control of what actually becomes a persistent instruction versus what was just context specific. It supports multiple destination files including project-specific rules, global configs, and even gitignored local preferences. The /reflect-skills command for discovering repeating patterns that could become skills is a nice touch for teams trying to systematize their workflows.
npx -y skills add bayramannakov/claude-reflect --skill claude-reflect --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