Z3r0 - AI-native red-team workbench for authorized penetr

项目说明

<p align="center">
<img src="assets/z3r0-logo.png" width="156" alt="Z3r0 logo" />
</p>

<p align="center">
<strong>English</strong>
</p>

<p align="center">
<a href="#architecture">Architecture</a> ·
<a href="#runtime-flow">Runtime flow</a> ·
<a href="#e">E model</a> ·
<a href="#sandbox-and-egress">Sandbox and egress</a> ·
<a href=" ·
<a href=" start</a>
</p>

<p align="center">
<strong>AI-nati workbench for authorized penetration testing and research, with specialist agents, sandboxed tooling, e records, and replayable timelines.</strong>
</p>

---

> :warning: Security notice
>
> This project is intended only for security testing, risk assessment, and academic research within legal and explicitly authorized scopes. It must not be used for unlawful, unauthorized, or destructi.
>
> This project does not grant permission to test, access, scan, or affect any third-party systems, networks, services, accounts, or data.
>
> The author is not responsible for any consequences, losses, damages, legal liabilities, or unlawful behavior caused by users.

O

Z3r0 is an AI-nati workbench for authorized penetration testing and research, with specialist agents, sandboxed tooling, e records, and replayable timelines. It combines a React operator console, a FastAPI management plane, a session-based multi-agent runtime, project-scoped e records, distributed Docker sandbox resources, and a controlled egress layer.

Z3r0 brings authorized scope, asset relationships, specialist assignments, e, findings, attack paths, workflow decisions, sandbox resources, and session timelines into one shared workspace. Red teams can coordinate execution, follow assessment progress, trace conclusions to supporting material, and review the complete operation without reconstructing state from separate con and tools.

Architecture

Z3r0 separates the system into four architectural planes:

| Plane | Scope |
| --- | --- |
| Control plane | Users, system configuration, agents, sessions, work projects, knowledge collections, managed hosts, sandbox images, sandbox containers, egress proxies, and optional FOFA asset-search integration. |
| Runtime plane | Multi-agent session execution, task-input LightRAG retrieval, li streaming, long-running task continuity, history projection, and timeline replay. |
| E plane | Authorized scope, asset relationships, graph-targeted work items, immutable e, findings, attack paths, target co, and workflow decisions. |
| Execution plane | Docker hosts, sandbox containers, shell/file/no, command execution, sandbox-local skills, built-in security tooling, and outbound network policy. |

This separation is reflected in the repository structure: routers and handlers expose application contracts, services own domain behavior, models define persistent state, and the React workbench consumes the stable REST/WebSocket surface.

Runtime flow

E model

Work projects provide a durable workspace for each assessment. Operators can explore the target en as an asset graph, assign specialists to specific assets and test surfaces, review the e produced by each assignment, and follow offensi through attack paths. Findings and path conclusions remain connected to their supporting material, giving the team a coherent record from initial scope through review and retesting.

| Data object | Role in the assessment |
| --- | --- |
| Work project | Assessment boundary for owners, authorized scope, sandbox bindings, sessions, workflow, and closure. |
| Asset | Canonically identified network, host, domain, service, application, endpoint, code, artifact, identity, data, or cloud entity. |
| Relation | E en assertion co structure, connectivity, dependencies, identity, trust, data flow, or pro. |
| Work item | Coordinated action unit connecting an assigned specialist with graph targets, dependencies, test surfaces, completion criteria, e, co conclusions, and lead review. |
| E | Immutable work item-attributed obser with pro, a stable source reference, supersession, and guarded in. |
| Finding | E security conclusion with , impact, disposition, and se kept consistent with optional C. |
| Attack path | Continuous sequence of offensi from entry to target, with e and optional ATT&CK mapping per step. |

The workbench gi a unified view of scope co, current assignments, blocked test surfaces, review queues, findings, attack paths, e, and specialist activity. Leads can review completed work, return specific target surfaces for further , and use focused search and filters to mo from project-le to the relevant asset, assignment, e chain, or security conclusion.

Sandbox and egress

Sandbox resources are managed infrastructure. Administrators manage Docker hosts, sandbox images, running containers, exposed ports, and project bindings. Operators and agents work through selected running containers, and the same sandbox boundary supports command execution, shell sessions, file management, browser/no, and sandbox-local skills.

The default sandbox image provides a focused security workspace with targeted DNS and ownership checks (dnsx, dig, nslookup, whois), low- HTTP inspection (curl, wget, httpx, openssl), focused service diagnostics (nc, nmap), local file and archi (file, sha256sum, 7z, unzip, tar, readelf), Android and firmware analysis (jadx, apktool, Ghidra, binwalk), binary and pwn tooling (gdb, Pwndbg, strace, ltrace, pwntools, and the pwntools-provided checksec), browser automation through agent-browser-cli, and Python workflows through uv.

Outbound traffic is normalized through a container-le profile. The sandbox runtime exports proxy en to a local proxy inside the container; the control plane can update the upstream policy to direct access or a managed HTTP, HTTPS, or SOCKS5 proxy. This gives the platform a unified place to manage network identity, traffic routing, and operator-en isolation.

Technical highlights

| Highlight | Description |
| --- | --- |
| Multi-agent orchestration | A lead agent coordinates specialist agents for intelligence gathering, , code audit, reverse analysis, and cryptanalysis. |
| Graph-dri | Work projects bind specialist work items to in-scope assets, test surfaces, e, findings, and attack-path . |
| Retrie plane | Building knowledge graphs with LightRAG Core provides matching original document chunks and graph context for task-oriented inputs. |
| Replayable event timeline | The UI consumes normalized timeline events that can be streamed live or loaded later as history. |
| Distributed sandbox resources | Managed Docker hosts, images, and containers allow execution en to be isolated, scaled, and assigned to projects. |
| Preloaded sandbox toolchain | The default image provides targeted DNS, HTTP, and service diagnostics plus local artifact, Android, firmware, reverse engineering, browser, and Python capabilities behind sandbox-local skills. |
| Unified egress layer | Container traffic can be routed through direct, HTTP, HTTPS, or SOCKS5 modes using one platform-managed policy surface. |
| Operator workbench | The frontend combines chat, workflow state, graph review, e chains, attack paths, sandbox selector, terminal, files, and noVNC. |

Expert team

| Code | Name | Role | Responsibilities |
| --- | --- | --- | --- |
| cso | Z3r0 | Chief security lead | Task decomposition, team coordination, result integration |
| cae | V3ra | Code audit engineer | Source code auditing, dependency review, remediation |
| cie | L1ly | Intelligence gathering engineer | Intelligence gathering, asset discovery, relationship mapping |
| cpe | Fr4nk | Penetration testing engineer | Penetration testing, , impact confirmation |
| cre | J4m3 | Reverse analysis engineer | Reverse analysis, firmware disassembly, binary unpacking |
| cce | Nu1L | Cryptography engineer | Cryptographic analysis, key review, security assessment |

Repository layout

Documentation

Acknowledgments

Thanks to the Linux.do website and its community for their support in project de and communication.

License

This project is licensed under the MIT License.

Star history

<a href="
<picture>
<source media="(prefers-color-scheme: dark)" srcset=" />
<source media="(prefers-color-scheme: light)" srcset=" />
<img alt="Star history chart" src=" />
</picture>
</a>

项目信息

开发语言：Python

Star 数：914

技术要点

该项目属于红蓝攻防方向的开源资源，代码结构清晰，适合安全研究人员学习参考。项目围绕AI-native red-team workbench for authorized penetration testing and vulnerability research, with specialist agents, sandboxed tooling, evidence records, and replayable timelines.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。