Activitywatch Analysis Skill

ActivityWatch Analysis Skill

A Claude Code skill for productivity analysis using [ActivityWatch](https://activitywatch.net/) data. Calculates focus scores, detects "death loops" (repetitive app switching), and generates actionable insights.

Features

• Calibration Mode: First-run setup that shows uncategorized apps, detects missing watchers, and guides personalization
• 80+ App Defaults: JetBrains IDEs, terminals, browsers, design tools, CRMs — works out of the box for most developers
• AI Agent Detection: Recognizes Claude Code, Codex, Aider, GitHub Copilot as productive workflows
• Dual Scoring: Productivity score (what you worked on) + Focus score (attention quality)
• Deep Browser Analysis: Site-level breakdown with productivity ratios (requires aw-watcher-web)
• Death Loop Detection: Identifies repetitive app switching patterns with fix suggestions
• Telegram Chat Categorization: Separate work chats from personal messaging
• Customizable Categories: JSON config to tune for your workflow

Requirements

• Python 3.8+
• ActivityWatch installed and running
• Recommended: pip install aw-client for direct API access
• Recommended: aw-watcher-web Chrome extension for site-level browser tracking

Install as Claude Code Skill

cp -r activitywatch-analysis-skill ~/.claude/skills/activitywatch-analysis

Quick Start

Step 1: Calibrate (first run)

# Show what needs configuring
python scripts/analyze_aw.py --fetch --from week --calibrate --config scripts/category_config.json

This shows:

• Uncategorized apps with significant usage time
• Whether aw-watcher-web is installed (critical for browser analysis)
• Telegram usage that may need chat-level categorization

Then edit `scripts/category_config.json` to classify your uncategorized apps.

Step 2: Analyze

# Today
python scripts/analyze_aw.py --fetch --from