TypeForge
A two-phase framework for synthesizing and selecting best-fit composite data types for stripped binaries using program analysis and LLM refinement.
0 projects
About TypeForge
TypeForge is a reverse engineering tool designed to recover composite data types, such as structures and unions, from stripped binaries. It operates in two distinct phases: a **Program Analysis phase** for initial synthesis and an **LLM-assisted Refinement phase** to improve accuracy. Currently implemented as a Ghidra extension, it supports headless mode for batch processing and single binary analysis. The tool includes components for data flow analysis, call graph generation, and type flow graph exploration. It also provides utilities for extracting ground truth from binaries containing debug symbols, though it requires DWARF-4 format as Ghidra does not currently support DWARF-5. While primarily a Ghidra extension, the project includes Python scripts for LLM-assisted double elimination and type inference wrapping. Development is ongoing for features such as direct result importation into Ghidra projects and a GUI mode.
Summarised from the official page linked above.