patch-diff-exploit

Pass

Audited by Gen Agent Trust Hub on Aug 2, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill requires executing high-privilege system commands such as dism, expand, and msiexec for extracting and manipulating Windows system patches. It also instructs the use of PowerShell with the Bypass execution policy to run local bootstrap scripts (e.g., bootstrap-reverse.ps1), which bypasses standard security restrictions.
  • [PROMPT_INJECTION]: The skill documents a workflow (references/root-cause-and-poc.md) where diff outputs and decompiled pseudocode from unvetted binaries are fed directly into an LLM for automated analysis. This ingestion of untrusted external data into the agent's context creates a surface for indirect prompt injection.
  • [EXTERNAL_DOWNLOADS]: The skill automates the retrieval of security tools and scripts from various external sources. It fetches components from GitHub repositories (google/bindiff, radareorg/radare2, joxeankoret/diaphora) and downloads kernel utility scripts from the official Linux kernel repository. It also retrieves PDB files directly from Microsoft's symbol servers.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 2, 2026, 12:44 PM
Security Audit — agent-trust-hub — patch-diff-exploit