From https://github.com/Sohan245/ReForge
ReForge is an end-to-end research framework that fuses AFL++ fuzzing, cloud-grade LLM exploit generation, and a local analyst model under a single orchestrator. At its core, ReForge pushes every crash- or hang-inducing input through an AI “forge”:
- AI-1 · Exploit Generator – turns raw crash data into a runnable Python PoC
- MCP Orchestrator – coordinates validation loops and state-sharing between the AIs
- AI-2 · Analyzer – produces a human-readable report that explains the exploit in plain English
The pipeline is model-agnostic — swap the cloud model, the local model, or the decision logic without touching the fuzzing harness. The result is a repeatable cycle that discovers, weaponises, and explains software weaknesses with minimal human intervention.
