Use this skill when the user says /performance, /digest, asks for a weekly Claude performance report, or wants to review how their Claude Code sessions have been going. Reads JSONL session files from ~/.claude/projects/, computes six effectiveness metrics (activity distribution, one-shot edit rate, subagent turn count, model mix, project allocation, hookify firings), fires diagnostic rules when thresholds breach, and writes behavioral rules into ~/.claude/CLAUDE.md so future sessions adapt. Closes the measurement loop for self-improving AI workflows.
2026-04-16