Claude Code Log Analyzer

Claude Code Log Analyzer

Python tools for analyzing Claude Code conversation history. Extract patterns, measure autonomous work hours, and understand your AI-assisted development workflow.

This toolkit was developed during a 97-day research project analyzing 543 hours of autonomous AI-assisted work across 165 releases. The methodology and findings are documented in the accompanying presentation.

What It Does

Three complementary analyses:

1. Arc Analysis - Identifies coherent units of work ("arcs") from your prompts. What kind of work are you doing?
2. Agent Analysis - Counts autonomous agent sessions and their durations. How much autonomous work happened?
3. Gate Analysis - Discovers and analyzes review gate tools in your sessions. How effective are your quality gates?

Tools:
- arc_analyzer.py - Detects work arcs using semantic analysis (Gemini Flash Lite), counts agent session hours
- chat_analyzer.py - Clusters prompts using Gemini embeddings + HDBSCAN to reveal interaction patterns
- gate_analyzer.py - Discovers review gate tools, extracts gate check results, classifies decisions and error types

Quick Start

Data Sources

The scripts read from your local Claude Code data directory (~/.claude/):

Custom Data Directory

To analyze logs from a different location (backup, server, another user):

Multiple Claude Directories

If you have Claude Code data in multiple locations (e.g., different user accounts on the same machine), set:

Arc Analyzer

Identifies "arcs" - coherent units of work that may span multiple prompts, tool uses, and agent spawns. Also counts autonomous agent session hours directly.

How Arc Detection Works

This tool uses semantic boundary detection with Gemini Flash Lite. For each user message, it asks: "Is this starting a new quest/journey, or…