Skip to content

Ghidrust provides a hyper fast, AI native reverse engineering platform built in Rust, inspired by Ghidra. Includes experimental GPU kernels to rapidly decompile large files, a decryption suite, and network monitor for packet to binary reverse engineering.

1.7 KB
Raw Blame

Stage-1 readability rubric

Human / semi-auto checklist for recovered C quality. Use alongside token_similarity from ghidrust ghidra-headtohead — Jaccard alone does not capture expression nesting or naming.

Scoring (per function)

Score each item 0 / 1. Report mean across a fixed function set.

# Check Pass when
1 Expression nesting Fewer single-use tN#v / reg#v temps than SSA ops; compound a + b * c present when arithmetic chains exist
2 Named calls Import / known function calls print as Name() not bare sub_<addr>() when the program has that symbol
3 Typed locals Stack locals declared with non-undefined types when evidence exists
4 Control flow Natural if / while / switch used; goto_rate for the function < 0.15 preferred
5 Early exit Early returns use return not goto L_ret when the target is a return block
6 OO (when RTTI) Methods show this / class pointer; virtual calls annotated when vtable known
7 Honesty No invented structs/enums; /* unimplemented */ left for unlifted ops

How to run

  1. Pick N functions from a fixture PE (or shared-entry head-to-head set).
  2. ghidrust decompile <bin> --addr <va> --json — inspect pseudo_c, folded_temps, goto_rate.
  3. Optional: ghidrust ghidra-headtohead … for token_similarity.
  4. Fill the table; track mean score over time. Do not claim Hex-Rays parity from this rubric alone.

Fixture themes (expand under fixtures/)

  • Expression-heavy arithmetic
  • Call-heavy (imports)
  • Loop / switch
  • Float/XMM (movss / movsd notes)
  • One medium public PE slice (no private install paths in docs)