Анализ сессий Claude через Braintrust

productivity

braintrust-analyze is a Claude Code skill that analyzes your working sessions using Braintrust tracing data, breaking down tool usage, agent spawns, skill activations, and token consumption across time. It runs via `uv run python -m runtime.harness scripts/braintrust_analyze.py` with flags like `--last-session`, `--agent-stats`, `--skill-stats`, `--detect-loops`, `--replay <session-id>`, and `--weekly-summary`. The loop detection feature flags sessions where the same tool was called more than five times consecutively, pointing to stuck search loops or inefficient approaches worth refactoring into new skills. Developers use it for post-task retrospectives, debugging failed sessions, and tracking weekly productivity trends. Requires a `BRAINTRUST_API_KEY` and the `braintrust-claude-plugin` for tracing to be enabled.