binary-diff

Pass

Audited by Gen Agent Trust Hub on Jul 17, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill documentation recommends installing the 'pyyaml' package via pip. This is a standard, well-known library for YAML processing.
  • [DATA_EXFILTRATION]: The skill transmits technical disassembly and pseudocode data to well-known LLM providers (OpenAI, Anthropic, and DeepSeek) for comparative analysis. This is documented as its primary function and targets established services.
  • [COMMAND_EXECUTION]: The skill instructions and associated Python script describe interacting with IDA Pro (via IDAPython or idapro_rename) to apply symbol migrations. This is standard behavior for reverse engineering automation tools.
  • [PROMPT_INJECTION]: The skill processes external binary data (disassembly/pseudocode) by interpolating it into LLM prompts.
  • Ingestion points: Untrusted disassembly and pseudocode are injected into the {disasm_code} and {procedure} variables in 'references/prompt-template.md'.
  • Boundary markers: The template uses triple backticks and markdown headers to delineate sections.
  • Capability inventory: The automation script utilizes 'httpx' for network requests and performs renaming operations within an IDA session.
  • Sanitization: There is no explicit sanitization of the input code strings before prompt interpolation, representing a standard indirect prompt injection surface inherent to LLM-based analysis tools.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 17, 2026, 09:25 AM
Security Audit — agent-trust-hub — binary-diff