If you're using Braintrust to trace your Claude Code sessions, this gives you a command line interface to actually make sense of that data. Run it after a complex task to see what tools you called, which agents spawned, and how many tokens you burned. The loop detection is genuinely useful because it'll catch when you're stuck calling the same tool repeatedly, which happens more than you'd think. Weekly summaries show your productivity patterns, and the skill/agent stats help you figure out what's actually working in your workflow. Requires your Braintrust API key and the tracing plugin already set up.
npx skills add https://github.com/parcadei/continuous-claude-v3 --skill braintrust-analyze