The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes an 'Agentic Review' feature designed to evaluate Simulink model guidelines by analyzing semantic data extracted from user-controlled models, which constitutes an indirect prompt injection surface.
Ingestion points: Model data is extracted via model_overview, model_read, and model_query_params as described in references/uncheckable-guidelines-review.md.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are defined for the extracted model content.
Capability inventory: The agent has access to evaluate_matlab_code and model_edit.
Sanitization: No sanitization of model strings is mentioned before evaluation.
[OBFUSCATION]: The skill includes several MATLAB P-code files (.p) which are obfuscated binary representations of the source code. These are standard vendor artifacts used to implement the skill's functionality.
[DYNAMIC_EXECUTION]: The skill executes its internal P-code scripts via the evaluate_matlab_code tool as part of its compliance analysis workflow.

checking-model-compliance