automotive-vehicle-security

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data sources which represents a potential attack surface.
  • Ingestion points: The skill ingests and processes CAN bus logs (SocketCAN/candump), DBC database files, and ECU firmware images using tools like cantools, binwalk, and Ghidra.
  • Boundary markers: The skill contains multiple sections dedicated to safety, legal considerations, and coordinated disclosure, explicitly warning the user to verify all operations and obtain written authorization.
  • Capability inventory: The skill utilizes shell commands (cansend, cangen, ip link) and Python scripts for network communication, file operations, and hardware interaction.
  • Sanitization: The provided examples rely on the user to validate data and do not include automated sanitization of log inputs.
  • [EXTERNAL_DOWNLOADS]: The skill references several external resources for tools and data, all of which are from trusted or well-known sources.
  • Fetches satellite ephemeris data from NASA's official CDDIS servers for GNSS spoofing simulations.
  • Downloads automotive databases and testing frameworks from established GitHub organizations such as Comma.ai, the Linux Foundation, and ChargePoint.
  • [COMMAND_EXECUTION]: The skill involves the execution of various system and automotive commands necessary for its stated purpose.
  • Uses sudo for system configuration tasks like loading kernel modules (vcan, slcan) and bringing up network interfaces (ip link).
  • Executes automotive-specific binaries (can-utils) and scripts to interact with Electronic Control Units (ECUs).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 04:36 PM