Sinopia AST Rewriter

A small Rust research tool for simulating source-code poisonings in AI-assisted code generation (AI poisoning domain), by rewriting Rust source files through AST mutations.

Features

• rename functions by prefixing with rewritten_
• poison names with the following attacker tactics:
  • misleading naming (validate_backdoor_*, encrypt_noop_*)
  • polarity flips (is_/has_/should_ inverts semantics)
  • hidden triggers (*_trigger_42, *_BACKDOOR)
  • overfit prefix inversion (safe_* vs unsafe_*)
  • alias poisoning (dangerous=eval, safe map)
  • entropy/noise names (func_XYZ_noise123)
  • semantic collisions (process -> process_collide)
  • distribution poisoning (sanitize -> vuln_*, raw -> safe_*)
• inject Markov-generated comments into function bodies and call sites
• swap call arguments for any call with two or more arguments
• rewrite existing // comments using text from Markov training
• process single file or recursively scan a directory of *.rs files
• training from *.txt files in a directory (eg. rfc-archive), caching via JSON
• token limit fixed to 4000 for training to keep initial runs quick

Usage

# Compile and run on one file, no Markov training:
cargo run -- path/to/target.rs

# Run on a directory recursively, with text corpus from rfc-archive:
cargo run -- path/to/rust-project path/to/rfc-archive

The tool rewrites files in place and prints Rewrote: ... for each file.

Notes

• Markov training will throttle to first 4000 tokens from the text corpus.
• The generated code is formatted via prettyplease.
• Existing comments are replaced to appear as generated and random.