The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions include fetching can-utils from the official linux-can GitHub repository and installing standard automotive security libraries such as python-can, udsoncan, scapy, and cantools via pip. These are well-known and established tools within the automotive research community.
[COMMAND_EXECUTION]: The skill provides numerous shell commands for low-level network operations, including modprobe for loading kernel modules, ip link for interface management, and candump/cansend for bus interaction. These are standard procedures for the described automotive analysis tasks.
[PRIVILEGE_ESCALATION]: The instructions utilize sudo for modprobe and ip link commands. This is a legitimate and necessary use of elevated privileges required to configure network interfaces and virtual CAN buses on Linux systems.
[INDIRECT_PROMPT_INJECTION]: The skill defines workflows for ingesting and parsing external data from CAN bus traffic and UDS diagnostic responses.
Ingestion points: CAN bus frames captured via candump and diagnostic responses received via udsoncan in Python scripts.
Boundary markers: The instructions explicitly advise the use of virtual CAN (vcan) interfaces for laboratory testing and emphasize physical isolation for real vehicle testing to prevent unintended consequences.
Capability inventory: The skill uses subprocess calls for can-utils, file-writing for logging (candump -l), and network operations via SocketCAN.
Sanitization: The analysis relies on standard parsers like cantools and udsoncan to interpret raw binary data based on predefined schemas (DBC files or UDS configs).

re-automotive