session-insights

You can analyze Claude Code session logs by applying a skill that processes recent JSONL files to extract message counts, tool usage, token consumption, and errors. It generates aggregated metrics, narrative insights, and classified friction categories.

To estimate token costs from Claude Code session histories, you can process up to 50 recent JSONL session logs. The analysis computes total token consumption and generates a report with estimated costs and high-cost prompt identification.

Yes, analyzing Claude Code sessions can surface repeated user instructions. The process scans JSONL logs to identify repetitive interactions and generates specific CLAUDE.md recommendations to automate or streamline those instructions.

The best way to audit developer sessions for tool failures is to scan JSONL session logs to extract and classify errors into taxonomy categories. This friction analysis identifies frequent tool failures and produces actionable recommendations.

Yes, session analysis skips interactions under three messages to ensure meaningful data extraction. This threshold filters out trivial exchanges in the JSONL logs, focusing the usage analytics on substantial developer sessions.

Yes, you can analyze git activity and file languages within Claude Code sessions. The process extracts git operations, most-worked directories, and file language signals from JSONL logs to detect usage patterns and surface activity metrics.