session-analyzer

Analyze a Claude Code or Claude Desktop agent-mode session transcript (JSONL) and produce a structured Markdown report. The report covers: all tool calls and their types, attribution of each action to a skill/agent rule vs. LLM autonomous decision, steps that caused errors and how the LLM recovered, total token usage and estimated cost, and optimization recommendations. Trigger when the user asks to "review", "analyze", "audit", or "report on" a Claude session, mentions a session directory path, a session UUID, an `audit.jsonl` / Claude Desktop log, or asks things like "what happened in that session", "what did Claude do in that run", "how many tokens did it use", "why did it retry", "what errors occurred", or "show me the tool calls". Also trigger when the user says "session report", "session summary", or references a session file by path or ID.