Claude Clue #30: Let Claude Grade Its Own Homework
Ask Claude to grade how well it's been helping you with the /insights command.
Claude Code will grade its own homework if you let it. Run /insights from any session, and it chews through the last 30 days of your Claude Code history, then opens an HTML report on how you actually work: how many sessions, which tools you lean on, what kinds of tasks you throw at it, and where things went sideways. Up top, there's an at-a-glance summary (what's working, what's hindering, quick wins, and bigger swings worth taking) sitting over an honest read of where Claude nailed it and where it fumbled.
The real gold is in the suggestions. Claude flags its own recurring mistakes, then hands you copy-paste-ready CLAUDE.md rules to stop them next time. It points at features you're probably underusing (MCP servers, custom skills, hooks, subagents) and splits its advice into quick tactical fixes and bigger strategic shifts to save you tokens and frustration. The friction section owns the ugly stuff too, misunderstood asks, buggy code, and all, which is oddly satisfying to read. It's a retrospective, except Claude runs it on itself (Haiku does the analysis).
Back in Clue #10 we ran a health checkup on your codebase. This points to the same idea in your collaboration: not whether the code is healthy, but whether you and Claude are actually working well together.
You can find out more information in their docs.
Claude reviewed its own work for this one. Grade pending.
HT: Roscoe Rawson