claude-code/README.md

# Claude Code v2.1.88 — Full Source

Source code extracted from `@anthropic-ai/claude-code@2.1.88` npm package via source maps.

The 57MB source map (`cli.js.map`) shipped inside the npm package and pointed to a Cloudflare R2 bucket containing the full TypeScript source. No reverse engineering was performed — Anthropic published it themselves.

---

## Quick Start

**Requirements:** Node.js 18+

```bash
git clone https://github.com/antonoly/claude-code.git
cd claude-code
node cli.js
```

On first launch, you'll be prompted to log in with your Anthropic account. If you have a Claude Max/Pro subscription, it works immediately — no API key needed.

### With an API Key

```bash
export ANTHROPIC_API_KEY=sk-ant-...
node cli.js
```

### With AWS Bedrock

```bash
export CLAUDE_CODE_USE_BEDROCK=1
export AWS_REGION=us-east-1
node cli.js
```

### With Google Vertex AI

```bash
export CLAUDE_CODE_USE_VERTEX=1
export CLOUD_ML_REGION=us-east5
node cli.js
```

### Custom API Endpoint

```bash
export ANTHROPIC_BASE_URL=https://your-proxy.example.com
export ANTHROPIC_API_KEY=your-key
node cli.js
```

### With Local Models (Ollama, LM Studio)

Claude Code uses the Anthropic Messages API format. To use local models, run [litellm](https://github.com/BerriAI/litellm) as a translation proxy:

```bash
# Terminal 1: Start litellm proxy
pip install litellm
litellm --model ollama/llama3.1:8b --port 8080

# Terminal 2: Point Claude Code at the proxy
export ANTHROPIC_BASE_URL=http://localhost:8080
export ANTHROPIC_API_KEY=not-needed
node cli.js
```

Works with any model Ollama supports — llama3.1, codellama, deepseek-coder, mistral, etc.

### With OpenAI / GPT Models

```bash
# Via litellm proxy
litellm --model openai/gpt-4o --port 8080

# Or any OpenAI-compatible endpoint (Codex, GPT-5.4, etc.)
litellm --model openai/o3 --port 8080
```

---

## What's Inside

| Stat | Value |
|------|-------|
| Lines of TypeScript | 512K |
| Source files | 1,923 |
| Built-in tools | 43 |
| Package version | 2.1.88 |

### Key Findings

- **`buddy/`** — Full Tamagotchi pet system: 18 species, rarity tiers (1% legendary, 0.01% shiny legendary), 5 stats, gacha PRNG seeded per-user, ASCII art sprites
- **`utils/undercover.ts`** — Auto-activates on public repos. Injects system prompt: *"Do not blow your cover."* No force-OFF switch
- **`outputStyles/`** — Anti-distillation defenses: strips tool lists, model info, and thinking content to prevent competitors from scraping behavior
- **`coordinator/`** — KAIROS/Auto-Dream: autonomous multi-step planning with self-reflection loops
- **`context/`** — 4-layer context compression: pruning → summarization → truncation → eviction

### Source Tree

```
├── cli.js              # Bundled CLI entry point (12MB)
├── package.json        # Original npm package metadata
├── vendor/             # Native binaries (ripgrep, audio-capture)
├── buddy/              # BUDDY pet system
├── tools/              # 43 built-in tools
├── commands/           # Slash commands
├── components/         # React/Ink terminal UI
├── context/            # Context window management
├── coordinator/        # Multi-agent orchestration
├── hooks/              # Pre/post tool execution hooks
├── plugins/            # Plugin/skill marketplace
├── services/           # API clients, auth, billing
├── skills/             # Built-in skills
├── utils/              # Undercover mode, anti-distillation, models
│   ├── undercover.ts   # "Do not blow your cover"
│   └── model/          # Model definitions, codename stripping
├── voice/              # Voice mode
└── vim/                # Vim keybindings
```

---

## Full Analysis

Independent architecture analysis with source code evidence:

**[verified-skill.com/insights/claude-code](https://verified-skill.com/insights/claude-code)**

---

## Disclaimer

This repository contains source code that was publicly exposed via npm source maps on March 31, 2026. All code was directly readable from the published npm package `@anthropic-ai/claude-code@2.1.88`. No reverse engineering, decompilation, or circumvention of technical protection measures was performed.