The Agent Skills Directory

[OBFUSCATION]: The file scripts/fileType.p is a MATLAB P-code file, which is a proprietary, obfuscated binary format.
The skill instructions require the agent to execute this binary script to perform file type verification.
Binary scripts cannot be statically audited for malicious behavior.
[DYNAMIC_EXECUTION]: The skill uses APIs that allow for the generation and runtime execution of arbitrary MATLAB code.
The setCallback function assigns MATLAB code strings to document events (e.g., AnalyzeFcn).
The setColumnFormula function assigns MATLAB code strings to spreadsheet columns for automatic calculation.
These features allow the agent to construct and embed executable logic within analysis documents.
[INDIRECT_PROMPT_INJECTION]: The skill defines a significant attack surface for indirect prompt injection by processing untrusted data to generate code.
Ingestion points: System structure and functional details are retrieved from Simulink models via model_overview and model_read, and from requirements text.
Boundary markers: There are no specified delimiters or instructions to ignore potential commands embedded within the model or requirement data.
Capability inventory: The skill can execute arbitrary MATLAB via evaluate_matlab_code and store persistent execution logic in documents through the dynamic execution features mentioned above.
Sanitization: The instructions lack requirements for sanitizing or escaping data from these external sources before incorporating it into documents or using it to build formulas.

manage-safety-analysis