Reflection

This is basically a post-mortem tool for your Claude sessions. Quick mode scans your chat history and CLAUDE.md to spot where Claude fumbled or could have been clearer, then helps you update the instructions to prevent repeat mistakes. Deep mode goes further with a full session breakdown covering problems solved, patterns discovered, user preferences, and knowledge gaps before proposing concrete updates. The implementation is thoughtful, it uses TodoWrite to track changes and waits for your approval before modifying anything. Honestly, this is the kind of meta-tooling that separates codebases that improve over time from ones that accumulate the same issues. If you're doing serious work with Claude Code, you'll want this in your rotation.

Install to Claude Code

npx -y skills add feiskyer/claude-code-settings --skill reflection --agent claude-code