The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes complex, user-supplied Simulink models and embedded MATLAB code which serves as an ingestion surface for potential indirect prompt injection attacks.
Ingestion points: The skill reads model structure, block parameters, signal names, and workspace variables using tools like model_read, model_query_params, and model_overview from .slx files and .sldd dictionaries.
Boundary markers: The instructions do not specify any boundary markers or delimiters to protect the agent's context when interpolating these potentially untrusted strings from the model into tool calls or reasoning steps.
Capability inventory: The skill possesses high-privilege capabilities via evaluate_matlab_code, which allows it to run arbitrary MATLAB code, modify the system environment, and perform file system operations.
Sanitization: While the skill's primary function is to sanitize models for external sharing, it does not implement sanitization for the agent's own input processing to prevent malicious data from being interpreted as instructions.
[DYNAMIC_EXECUTION]: The skill dynamically generates and modifies executable content during the sanitization process.
It is instructed to create anonymized .m stub files for external function dependencies that are not available on the path.
It modifies script content within MATLAB Function blocks and generates an anonymized version of the model that is then executed (compiled) in Step 8 to verify functionality.
[COMMAND_EXECUTION]: The skill extensively uses command execution through the evaluate_matlab_code tool to perform its operations.
It executes shell-like operations within the MATLAB environment to rename blocks, clear metadata, and manage the model lifecycle (e.g., bdclose, copyfile, and SLDD management).
These commands are used to implement the core functionality of the skill but involve significant interaction with the host environment.

simulink-sanitize-model