The Agent Skills Directory

[OBFUSCATION]: The skill utilizes 11 MATLAB P-code files (e.g., canHardwareDiscovery.p, canSupportSummary.p) which contain obfuscated, non-human-readable code. While P-code is a standard MATLAB format for distributing protected code, its use here prevents a full security audit of the diagnostic logic.\n- [DYNAMIC_EXECUTION]: The agent is instructed to run the obfuscated scripts via the evaluate_matlab_code tool. The documentation in references/can/troubleshooting/diagnostic-workflow.md explicitly tells the agent to "Run functions, don't read them," which bypasses the potential for the agent to verify the scripts' behavior before execution.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze vehicle network communication (CAN/CAN FD messages) from live buses or log files. This untrusted external data is processed by diagnostic scripts, and the results are returned to the agent's context, creating a surface for indirect prompt injection if the processed data contains malicious instructions.\n
Ingestion points: Untrusted data enters the agent context through the receive function (documented in references/can/receive.md) and the analysis pipeline described in references/can/troubleshooting/l2-analysis-pipeline.md.\n
Boundary markers: The instructions lack explicit boundary markers or warnings for the agent to ignore instructions embedded within the network traffic data.\n
Capability inventory: The skill has access to the evaluate_matlab_code tool, which can execute MATLAB commands, and it can transmit data onto a physical vehicle network via the transmit functions.\n
Sanitization: There is no evidence of sanitization or filtering of the content within the CAN message payloads before they are presented to the agent for analysis.

matlab-use-vehicle-network