Share
## https://sploitus.com/exploit?id=0725F40C-1A72-5C8A-8B5C-E63D3AC478E8
# Automation-Exploit (Legacy Prototype)

This repository contains an **earlier prototype** of *Automation‑Exploit*, a multi‑agent framework designed to automate reconnaissance-driven exploit generation using LLM-based reasoning and orchestrated analysis pipelines.

⚠️ **Note**  
This is **not** the latest version of the framework.  
The complete architecture — including the **Digital Twin safety layer**, **double-container execution model**, **Autopsy mechanism**, **Accumulator-based MAS**, and the **full adversarial governance pipeline** — is part of a newer release (**v3**) currently under **peer review for publication in *Computers & Security (Elsevier)***.  
These components will be released only after the review process is completed.

---

## 🚀 Overview (Legacy Version)

This legacy version includes:

- The original **orchestration pipeline** for coordinating LLM agents  
- Early-stage **exploit generation logic**  
- Basic **contextual intelligence extraction**  
- Initial **multi-step reasoning workflow**  

- A preliminary **Digital Twin prototype** based on single-container mirroring  
  (no Predictive Safety Layer, no dual-container replication, no deterministic replay, no Autopsy system)
- Early **Adversarial Hand-off** mechanism for delegating exploit refinement between agents  
  (without accumulator-driven meta-reasoning)
- A basic **Two-Stage False-Positive Reduction** workflow  

- Prototype modules for:
  - reconnaissance interpretation  
  - vulnerability pattern inference  
  - exploit template adaptation  

This prototype represents the foundation on which the more advanced architecture was built.

---

## 🧠 What’s *not* included (by design)

The following components are **not part of this public prototype**, as they belong to the version currently under peer review:

- **Digital Twin–based execution environment**  
- **Double-container safety mechanism** (crash isolation + deterministic replay)
- **Isomorphic Synchronization & LD_PRELOAD-based Replication**  
  (runtime hooking, FD interception, deterministic state alignment, and the 7‑stage autonomous exfiltration pipeline) 
- **Autopsy subsystem** for post-crash forensic telemetry  
- **Accumulator-driven Multi-Agent System (MAS)**  

These features introduce novel scientific contributions and will be released only after the publication process.

## 📄 Citation (Preprint)

The official preprint of this research is publicly available on **arXiv**:

**Biagio Andreucci, Arcangelo Castiglione.  
"Automation-Exploit: A Multi-Agent LLM Framework for Adaptive Offensive Security with Digital Twin-Based Risk-Mitigated Exploitation."  
arXiv:2604.22427 [cs.CR] (2026).**  
https://arxiv.org/abs/2604.22427

---

## 📬 Contact

For research discussions or collaborations:  
**Biagio Andreucci**  
AI & Cybersecurity Engineer  
LinkedIn: https://www.linkedin.com/in/biagio-andreucci-42672627a

---

## ⚖️ Disclaimer

This repository is provided **for research purposes only**.  
No real-world exploitation should be performed using this prototype.