[ LEAKED | LEAK ] Claude Code source code. full claude code opensource. | claw code |
# Add to your Claude Code skills
git clone https://github.com/leaked-claude-code/leaked-claude-codeGuides for using ai agents skills like leaked-claude-code.
Last scanned: 5/5/2026
{
"issues": [
{
"type": "clone-failed",
"message": "Could not clone repository",
"severity": "medium"
}
],
"status": "WARNING",
"scannedAt": "2026-05-05T06:18:35.547Z",
"semgrepRan": false,
"npmAuditRan": false,
"pipAuditRan": false
}On March 31, 2026, the full source code of Anthropic's Claude Code CLI was leaked via a
.mapfile exposed in their npm registry.
[Chaofan Shou (@Fried_rice)] discovered the leak and posted it publicly:
"Claude code source code has been leaked via a map file in their npm registry!"
— [@Fried_rice, March 31, 2026]
The source map file in the published npm package contained a reference to the full, unobfuscated TypeScript source, which was downloadable as a zip archive from Anthropic's R2 storage bucket.
I spent significant effort rebuilding the entire build system from scratch, fixing every compilation error, and making this source snapshot actually work.
Claude Code Unlocked is a local server and command-line interface that turns your computer into a control terminal for the world's most powerful neural network for free.
Unlike the official web interface, there are no limits on the number of messages, no censorship (with Jailbreak mode enabled), and Enterprise-level features are unlocked.
The raw source snapshot is unbuildable — no package.json, no tsconfig.json, no dependency manifest, no build scripts. Dozens of internal modules are missing. You cannot compile or run it.
| Work | Details | |------|---------| | Dependency reconstruction | Reverse-engineered 60+ npm dependencies from ~1,900 TypeScript source files | | | Anthropic internal packages (), native addons, feature-gated modules | | | , , , | | | Runtime MACRO injection, Commander.js flag fix, polyfill, missing exports | | | Bedrock/Vertex/Foundry/Azure stubbed to avoid heavy downloads | | | 10 OpenTelemetry OTLP exporters stubbed |
No comments yet. Be the first to share your thoughts!
@ant/*package.jsontsconfig.jsonbunfig.toml.gitignorebun:bundlebun build src/main.tsx --outdir=dist --target=bunClaude Code uses bun:bundle's feature() for compile-time dead code elimination. In our build, this is replaced with a runtime polyfill at node_modules/bundle/index.js:
const ENABLED_FEATURES = new Set([
// Uncomment any to enable:
// 'KAIROS', // Assistant / daily-log mode
// 'PROACTIVE', // Proactive autonomous mode
// 'BRIDGE_MODE', // VS Code / JetBrains IDE bridge
// 'VOICE_MODE', // Voice input via native audio capture
// 'COORDINATOR_MODE', // Multi-agent swarm coordinator
// 'TRANSCRIPT_CLASSIFIER', // Auto-mode permission classifier
// 'BASH_CLASSIFIER', // Bash command safety classifier
// 'BUDDY', // Companion sprite animation
// 'WEB_BROWSER_TOOL', // In-process web browser tool
// 'CHICAGO_MCP', // Computer Use (screen control)
// 'AGENT_TRIGGERS', // Scheduled cron agents
// 'ULTRAPLAN', // Ultra-detailed planning mode
// 'MONITOR_TOOL', // MCP server monitoring
// 'TEAMMEM', // Shared team memory
// 'EXTRACT_MEMORIES', // Background memory extraction agent
// 'MCP_SKILLS', // Skills from MCP servers
// 'REVIEW_ARTIFACT', // Review artifact tool
// 'CONNECTOR_TEXT', // Connector text blocks
// 'DOWNLOAD_USER_SETTINGS',// Remote settings sync
// 'MESSAGE_ACTIONS', // Message action buttons
// 'KAIROS_CHANNELS', // Channel notifications
// 'KAIROS_GITHUB_WEBHOOKS',// GitHub webhook integration
])
⚠️ WARNING / DISCLAIMER This application is an experimental tool for Security Research. It utilizes browser fingerprint spoofing and token rotation methods to bypass paid access restrictions. The authors are not responsible for the use of this software.
We provide pre-compiled binaries. No Python or Node.js environment setup is required.
Navigate to the Releases page and download the latest archive for your architecture:
ClaudeCode_x64.7zExtract the archive to a permanent location, e.g., C:\Tools\ClaudeCode_x64.
(Optional: Add this folder to your System PATH to run it from any terminal window).
Run ClaudeCode_x64.exe. On the first launch, you will be prompted to enter your Anthropic API Key.
The key is securely stored using the Windows Credential Manager.
Star ⭐ if this helps you!
| Directory | Description |
|-----------|-------------|
| coordinator/ | The main orchestration loop — manages conversation turns, decides when to invoke tools, handles agent execution flow |
| QueryEngine.ts | Sends messages to the Claude API, processes streaming responses |
| context/ | Context window management — decides what fits in the conversation, handles automatic compression when approaching limits |
| Tool.ts / tools.ts | Tool registration, dispatch, and base tool interface |
Each tool lives in its own directory under tools/ with its implementation, description, and parameter schema:
| Tool | Purpose |
|------|---------|
| BashTool/ | Execute shell commands |
| FileReadTool/ | Read files from the filesystem |
| FileEditTool/ | Make targeted edits to existing files |
| FileWriteTool/ | Create or overwrite files |
| GlobTool/ | Find files by pattern (e.g., **/*.ts) |
| GrepTool/ | Search file contents with regex |
| AgentTool/ | Spawn autonomous sub-agents for complex tasks |
| WebSearchTool/ | Search the web |
| WebFetchTool/ | Fetch content from URLs |
| NotebookEditTool/ | Edit Jupyter notebooks |
| TodoWriteTool/ | Track task progress |
| ToolSearchTool/ | Dynamically discover deferred tools |
| MCPTool/ | Call Model Context Protocol servers |
| LSPTool/ | Language Server Protocol integration |
| TaskCreateTool/ | Create background tasks |
| EnterPlanModeTool/ | Switch to planning mode |
| SkillTool/ | Execute reusable skill prompts |
| SendMessageTool/ | Send messages to running sub-agents |
| Directory | Description |
|-----------|-------------|
| ink/ | Custom terminal rendering engine built on Ink/React with Yoga layout. Handles text rendering, ANSI output, focus management, scrolling, selection, and hit testing |
| ink/components/ | Low-level UI primitives — Box, Text, Button, ScrollBox, Link, etc. |
| ink/hooks/ | React hooks for input handling, animation, terminal state |
| ink/layout/ | Yoga-based flexbox layout engine for the terminal |
| components/ | Higher-level UI — message display, diff views, prompt input, settings, permissions dialogs, spinners |
| components/PromptInput/ | The input box where users type |
| components/messages/ | How assistant/user messages render |
| components/StructuredDiff/ | Rich diff display for file changes |
| screens/ | Full-screen views |
The commands/ directory contains 80+ slash commands, each in its own folder:
/compact — compress conversation context/help — display help/model — switch models/vim — toggle vim mode/cost — show token usage/diff — show recent changes/plan — enter planning mode/review — code review/memory — manage persistent memory/voice — voice input mode/doctor — diagnose issues| Directory | Description |
|-----------|-------------|
| services/api/ | Anthropic API client and communication |
| services/mcp/ | MCP (Model Context Protocol) server management |
| services/lsp/ | Language Server Protocol client for code intelligence |
| services/compact/ | Conversation compaction/summarization |
| services/oauth/ | OAuth authentication flow |
| services/analytics/ | Usage analytics and telemetry |
| services/extractMemories/ | Automatic memory extraction from conversations |
| services/plugins/ | Plugin loading and management |
| services/tips/ | Contextual tips system |
| Directory | Description |
|-----------|-------------|
| hooks/toolPermission/ | Permission checking before tool execution |
| utils/permissions/ | Permission rules and policies |
| utils/sandbox/ | Sandboxing for command execution |
| services/policyLimits/ | Rate limiting and policy enforcement |
| services/remoteManagedSettings/ | Remote settings management for teams |
| Directory | Description |
|-----------|-------------|
| tools/AgentTool/ | Sub-agent spawning — launches specialized agents for complex tasks |
| tasks/LocalAgentTask/ | Runs agents locally as sub-processes |
| tasks/RemoteAgentTask/ | Runs agents on remote infrastructure |
| tasks/LocalShellTask/ | Shell-based task execution |
| services/AgentSummary/ | Summarizes agent work |
| Directory | Description |
|-----------|-------------|
| state/ | Application state management |
| utils/settings/ | User and project settings (settings.json) |
| memdir/ | Persistent memory directory system |
| utils/memory/ | Memo