The Agent Skills Directory

[OBFUSCATION]: The skill distributes its primary functional logic using five MATLAB P-code (.p) files located in the scripts/ directory. P-code is a protected, non-human-readable format that obfuscates the actual MATLAB instructions, preventing security verification of the operations performed by these scripts when invoked by the agent.
[DYNAMIC_EXECUTION]: The skill relies on the mcp__matlab__evaluate_matlab_code tool to dynamically execute strings of MATLAB code. This includes invoking the obfuscated scripts and processing user-provided parameters, creating a broad attack surface for arbitrary code execution within the connected MATLAB environment.
[COMMAND_EXECUTION]: The skill performs direct file system operations via the radarDesignerSession wrapper. It supports 'save' and 'load' actions for MATLAB session files (.mat), allowing the agent to read from and write to the host file system based on potentially untrusted user input.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection as it processes structured output from an external application to present to the user.
Ingestion points: The skill ingests data from the Radar Designer app via radarDesignerResults(h, 'read'), which returns a table of performance metrics.
Boundary markers: None; there are no instructions to the agent to treat the data returned from the application as untrusted or to ignore embedded instructions.
Capability inventory: The skill has significant capabilities including arbitrary MATLAB code evaluation and file system write access.
Sanitization: There is no evidence of validation or sanitization of the application's output before the agent processes and summarizes it.

matlab-design-radar