The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes system-level commands to discover available CAN hardware interfaces.
Evidence: scripts/can_runtime.py and scripts/can_scan.py use subprocess.run to call powershell, lsusb, and ip for enumerating USB devices and network interfaces.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources (CAN bus traffic and log files), presenting a surface for indirect prompt injection.
Ingestion points: Data is ingested from live CAN interfaces in can_monitor.py, can_log.py, and can_stats.py, and from external files in can_decode.py.
Boundary markers: No explicit boundary markers or safety instructions are used when displaying the content of CAN messages to the agent.
Capability inventory: The skill has the ability to write to the CAN bus using python-can, write files to the local system (logs), and execute system discovery commands.
Sanitization: CAN data is treated as hex strings or raw bytes, which minimizes but does not eliminate the risk of the LLM interpreting data as instructions.

can