Code Wrapped

Spotify Wrapped, but for your AI pair programming sessions.
What did you really build this year?

---

You've been pair programming with AI all year. Thousands of prompts. Countless debugging sessions at 2am. That mass-refactoring spree in October.

Code Wrapped parses your local session history and turns it into something you can actually look at (and maybe share, if you're brave).

What You'll Discover

• Your coding archetype — Are you The Debugger? The Architect? The 3am Explorer?
• Your vibe breakdown — How much time in "flow state" vs "debugging hell"?
• Your tool DNA — Terminal warrior or edit-everything type?
• Awards you didn't know you earned — Night Owl, Marathon Coder, Repo Hopper...
• The numbers — Sessions, turns, tokens consumed, streaks

All from data that's already on your machine. Nothing leaves your computer unless you want it to.

Quick Start

# Clone it
git clone https://github.com/cipher982/code-wrapped.git
cd code-wrapped

# Install (requires uv)
uv sync

# Generate your 2025 wrapped
uv run code-wrapped run --year 2025

Opens data/output/wrapped-2025.html in your browser. That's it.

Options

# Add AI-generated narrative commentary
uv run code-wrapped run --year 2025 --narrate

# Just the JSON, skip the pretty stuff
uv run code-wrapped run --year 2025 --no-report

# Verbose mode (see what's being parsed)
uv run code-wrapped run --year 2025 -v

What Gets Generated

data/output/
├── wrapped-2025.html          # Interactive report with charts
├── wrapped-2025.json          # Full stats (keep this private)
├── wrapped-2025-share.json    # Safe to share version
└── cards/
    ├── hero-stats.png         # Your big numbers
    ├── tool-fingerprint.png   # Your coding DNA
    ├── agent-comparison.png   # Claude vs Codex vs...
    └── award-*.png            # Your achievements

The PNG cards are sized for Twitter/social sharing (1200x630).

Supported AI Agents

Agent	Data Location	What's Parsed
Claude Code	~/.claude/projects/	Sessions, tools, tokens, timestamps
Codex	~/.codex/sessions/	Sessions, messages, git context
Cursor	~/Library/.../state.vscdb	Composer sessions, modes
Gemini	~/.gemini/tmp/	Sessions, messages

Don't use all of them? No problem — it'll work with whatever it finds.

The Analytics

Topics Detected

API integrations, frontend, backend, database, DevOps, testing, documentation, refactoring, debugging, AI/ML, mobile, security, performance, data processing

Vibes Classified

• Debugging Hell — errors, stack traces, "why won't this work"
• Flow State — shipping, completing, "it works!"
• Learning Mode — "how does X work", exploring docs
• Deep Work — long sessions, complex problems
• Exploration — trying things, prototyping

Archetypes Assigned

• The Architect — designs, structures, refactors
• The Debugger — fixes, investigates, traces
• The Explorer — asks "how", learns, experiments
• The Builder — creates, implements, adds features
• The Shipper — deploys, releases, pushes
• The Tester — validates, checks, covers

Awards Earned

Night Owl, Early Bird, Marathon Coder, Speed Demon, Streak Master, Repo Hopper, Deep Diver, AI Whisperer, Polyglot, Weekend Warrior, Terminal Master, Bug Slayer

Privacy

Your data stays local. The generated files in data/output/ contain your personal stats — the .gitignore excludes them by default.

The --narrate flag sends aggregated stats (not code, not prompts) to the Anthropic API to generate commentary. Requires ANTHROPIC_API_KEY.

Requirements

• Python 3.11+
• uv (recommended) or pip
• Session data from at least one supported AI agent

Development

# Install with dev dependencies
uv sync --group dev

# Run tests
uv run pytest

# Lint
uv run ruff check .

How It Works

1. Parse — Reads JSONL/SQLite session files from each agent's data directory
2. Normalize — Converts to unified Session model with common fields
3. Enrich — Detects topics, vibes, archetypes via keyword matching
4. Analyze — Computes streaks, distributions, records, awards
5. Generate — Creates HTML report with Plotly charts + PNG cards with Pillow
6. Narrate (optional) — Sends context to Claude for Wrapped-style commentary

FAQ

Q: Why not just use the built-in stats from Claude/Cursor/etc? A: They don't exist. Also, cross-agent comparison is fun.

Q: Is this official/affiliated with Anthropic/OpenAI/etc? A: No. It just reads the local files these tools create.

Q: Can I share my wrapped publicly? A: The wrapped-{year}-share.json strips sensitive paths. The PNG cards are designed for sharing. The full JSON has everything — keep that private.

Q: It says I have 0 sessions? A: Check that your AI tools store data in the expected locations. Run with -v to see what's being scanned.

License

MIT

---

Built during a mass debugging session at 1am. The irony is not lost.