Session Review Skill

Generate a structured retrospective report for the current Claude Code session.

Core Principle

Extract only human-readable content. This means:

• ✅ User messages (the human's prompts/questions/instructions)
• ✅ Agent prose responses (explanations, reasoning, summaries, answers)
• ❌ Tool calls (bash commands, file reads/writes, search queries)
• ❌ Tool results (command output, file contents, API responses)
• ❌ System messages and internal metadata

Extraction Script

A companion Python script handles robust extraction from session JSONL files: