claude-leaked-files - Mirrored snapshot of Claude Code's source (exposed

项目说明

Claude Code Source Snapshot (For Security Research)

This repository mirrors a publicly exposed snapshot of Claude Code’s source that became accessible on March 31, 2026, due to a source map exposure in the npm package.

It is preser for:
* Educational purposes
* Defensi research
* Software supply-chain analysis

> [!CAUTION]
> This repository does not claim ownership of the original code and is not an official Anthropic repository.

---

Research Background

This repository is maintained by a uni student researching:
* Software supply-chain leaks and build artifact exposure
* Secure software engineering practices
* Architecture of agent-based de tools
* Defensi of real-world CLI systems

The archi:
* Academic study
* Hands-on security research
* Architectural review
* Discussion of packaging and release-process failures

---

How the Source Became Public

On March 31, 2026, Chaofan Shou () reported that Claude Code source files were accessible through a .map file included in the npm distribution.

> “Claude code source code has been leaked via a map file in their npm registry!”
> — , March 31, 2026

The source map pointed to unobfuscated TypeScript source files hosted in Anthropic’s R2 storage bucket, making the entire src/ directory publicly downloadable.

---

What This Repository Contains

Claude Code is Anthropic’s command-line interface for interacting with Claude to perform software engineering tasks such as:
* Editing files
* Running shell commands
* Searching codebases
* Managing workflows
* Coordinating multiple agents

Snapshot details:

* Public exposure date: 2026-03-31
* Language: TypeScript
* Runtime: Bun
* Terminal UI: React + Ink
* Size: ~1,900 files, 512,000+ lines of code

---

Directory Structure O

Architecture O

1. Tool System (src/tools/)

Each tool Claude Code can use is implemented as a self-contained module. Every tool defines its input schema, permission requirements, and execution logic.

| Tool | Purpose |
| :--- | :--- |
| BashTool | Run shell commands |
| FileReadTool | Read files (text, images, PDFs, notebooks) |
| FileWriteTool | Create or o files |
| FileEditTool | Partial file edits |
| GlobTool | File pattern search |
| GrepTool | Content search (ripgrep) |
| WebFetchTool | Fetch URL content |
| WebSearchTool | Web search |
| AgentTool | Spawn sub-agents |
| SkillTool | Execute skills |
| MCPTool | MCP ser |
| LSPTool | Language Ser |
| NotebookEditTool | Jupyter notebook edits |
| TaskCreate / Update | Task management |
| SendMessageTool | Inter-agent messaging |
| TeamCreate / Delete | Team management |
| PlanMode tools | Planning mode toggle |
| Worktree tools | Git worktree isolation |
| CronCreateTool | Scheduled triggers |
| RemoteTriggerTool | Remote triggers |
| SleepTool | Wait / proactive mode |
| SyntheticOutputTool | Structured output |

2. Command System (src/commands/)

Slash commands (prefixed with /) provide user-facing functionality.
* /commit – create git commits
* /review – code review
* /compact – context compression
* /config – settings
* /doctor – diagnostics
* /login / /logout – authentication
* /memory – persistent memory
* /skills – skill management
* /tasks – task tracking
* /vim – Vim mode
* /diff – view changes
* /cost – usage cost
* /resume – restore session
* /share – share session
* /desktop / /mobile – app handoff

3. Service Layer (src/services/)

Handles integrations and system-le:
* Anthropic API client
* MCP ser
* OAuth 2.0 authentication
* LSP management
* Analytics and feature flags
* Plugin loading
* Context compression
* Policy enforcement
* Token estimation
* Team memory syncing

4. Bridge System (src/bridge/)

Provides two-way communication between the CLI and IDEs (VS Code, JetBrains).
* Message protocol
* Permission callbacks
* REPL bridge
* JWT authentication
* Session execution control

5. Permission System

Every tool in is checked through the permission system. Permissions may be prompted to the user, auto-approved, denied, or bypassed.
Modes: Default, Plan Mode, Auto, and BypassPermissions.

6. Feature Flags

Uses Bun’s bun:bundle feature flags to completely remo code at build time.
Notable flags: PROACTIVE, KAIROS, BRIDGE_MODE, DAEMON, , AGENT_TRIGGERS, MONITOR_TOOL.

---

Important Files

* QueryEngine.ts (~46K lines): Core LLM engine handling streaming responses, tool-call loops, retries, token tracking, and thinking mode.
* Tool.ts (~29K lines): Defines base tool interfaces, schemas, permissions, and progress states.
* commands.ts (~25K lines): Registers and executes all slash commands with en loading.
* main.tsx: CLI startup logic including command parsing (Commander.js), Ink UI initialization, and parallel prefetching.

---

Technology Stack

| Area | Technology |
| :--- | :--- |
| Runtime | Bun |
| Language | TypeScript (strict) |
| UI | React + Ink |
| CLI | Commander.js |
| | Zod v4 |
| Search | ripgrep |
| Protocols | MCP, LSP |
| API | Anthropic SDK |
| Telemetry | OpenTelemetry + gRPC |
| Feature Flags | GrowthBook |
| Auth | OAuth 2.0, JWT, macOS Keychain |

---

Design Highlights

* Parallel Prefetching: Startup performance is improved by loading MDM settings, keychain data, and API connections in parallel before hea.
* Lazy Loading: Large dependencies (telemetry, analytics, gRPC) are dynamically imported only when needed.
* Agent Swarms: Multiple agents can run in parallel using AgentTool and the coordinator system.
* Skill System: Reusable workflows live in skills/ and can be extended by users.
* Plugin Architecture: Supports both built-in and third-party plugins.

---

Disclaimer

This repository is an educational and defensi research archi by a uni student. Its purpose is to study source exposure incidents, packaging and release failures, and modern agent-based CLI architectures.

All original Claude Code source remains the property of Anthropic. This repository is not affiliated with, endorsed by, or maintained by Anthropic.

项目信息

开发语言：TypeScript

Star 数：285

技术要点

该项目属于防护思路方向的开源资源，代码结构清晰，适合安全研究人员学习参考。项目围绕Mirrored snapshot of Claude Code's source (exposed 2026-03-31) preserved for educational purposes, defensive security research, and software supply-chain analysis.展开，提供了可运行的实现，对理解相关安全机制、工具原理与攻防技术具有较高的研究价值。

从技术实现角度看，此类项目的核心在于对底层原理的深入理解与工程化落地。开发者在研究过程中可以结合官方文档与社区讨论，梳理项目的初始化流程、核心模块职责与数据流转，逐步掌握其中的技术细节与设计思路。

使用与学习建议

1. 下载源码压缩包，解压到本地环境。

2. 根据项目类型准备对应的运行环境与依赖。

3. 按照 README 中的步骤完成配置与部署。

4. 在隔离的测试环境中运行验证，避免影响生产系统。

免责声明

本资源来源于开源社区，仅供安全研究、技术学习与合法授权范围内的攻防演练使用，请勿用于任何非法或侵害他人权益的用途。使用本项目产生的一切后果由使用者自行承担。