session-reflection-analysis

To analyze JSONL session logs for token waste, use jq to generate a token-reduced summary of your Claude Code history. This process detects repeated mistakes, missing automation, and context-loss after compaction without reading raw session files directly.

Finding repeated mistakes requires a session-level reflection that reviews recent JSONL session data. By summarizing the last 7 days of history, you can identify slow iterations, such as re-reading materials or re-implementing existing features, and flag them for correction.

You generate copy-paste-ready workflow improvement proposals by transforming session reflection findings into structured recommendations. These proposals cover CLAUDE.md updates, new skills, slash commands, and scripts to automate previously inefficient development tasks.

Yes, jq is a required dependency to extract and summarize JSONL session data. It processes the raw project logs to create a token-reduced summary, enabling the analysis to detect context-loss and inefficiencies without directly reading the original files.

Preventing context loss after compaction involves identifying persistent context that should exist in documentation. Session analysis flags re-derived context and recommends specific documentation updates to ensure critical information survives future compaction events.

Session analysis produces structured proposals to update CLAUDE.md and create new slash commands, but it does not apply them automatically. It outputs dated reflection documents with high, medium, and low priority recommendations for you to copy and paste.