If you use Claude Code regularly, there's a built-in command that's easy to miss: /insights. It analyses your local session history and reports on how you actually work.
What it does
Run it inside any Claude Code session:
/insightsClaude Code scans your recent sessions on the machine and generates an interactive HTML report covering things like:
- Which projects you spend the most time in
- Which commands and features you reach for most
- Recurring patterns in how you structure prompts
- Friction points, like Claude claiming success without verifying, or suggesting a tool outside your stack
- Context window usage trends over time
Why it's useful
It's a personal, local report on how you work, not a usage dashboard for your org. It looks across weeks of your own sessions and surfaces habits you wouldn't otherwise notice: prompts you rewrite constantly, corrections you keep making, or friction that shows up every time you touch a specific part of the codebase.
That data is actionable in a very specific way. Once you've got a report:
- Fold recurring corrections into your
CLAUDE.mdso Claude stops making the same mistake - Turn multi-step workflows you repeat often into custom slash commands
- Flag the sessions with the most friction and figure out if the prompt, not the model, was the problem
Using it well
/insights is retrospective, so it's most useful once you have a few weeks of session history built up. Run it, make a couple of concrete changes to your config, then run it again in a few weeks to see if the friction points actually went away. Treated as a one-off curiosity it's a neat report; treated as a recurring check-in, it's closer to a coach for how you work with AI.