lessons-extractor

You can extract lessons from Claude Code session logs by automating the summarization, extraction, and merging of session data. This process identifies tasks, decisions, and outcomes to generate structured, reusable lessons from your project experiments.

The best way to summarize Claude Code sessions and track tool failures is to use an automated extraction process that identifies patterns across multiple sessions. This allows you to capture end-to-end outcomes and turn them into structured lessons.

Yes, you can generate both human-readable Markdown and machine-readable JSONL outputs for your AI lessons. This dual-output approach allows developers to read insights directly while enabling downstream automated tools to parse the data efficiently.

Yes, the lessons extraction process supports incremental runs with run-scope guidance to avoid reprocessing old logs. It also includes data redaction and deduplication features to maintain high-quality, secure insights across multiple runs.

The extracted lessons and summaries are stored under the docs/ai/lessons-extractor directory in your project. This centralizes the human-readable lessons.md and machine-readable lessons.jsonl outputs for easy team access.