The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes external data from Simulink models (.slx) and data dictionaries (.sldd).
Ingestion points: Data enters the agent context via model_read and model_query_params when inspecting model configurations and dictionary entries (e.g., in references/code_mapping.md and references/coder_dictionary.md).
Boundary markers: The skill documentation mandates calling introspect_api to retrieve allowed values and properties, which acts as a runtime constraint, although explicit delimiters for external content are not defined.
Capability inventory: The skill can execute persistent changes to the filesystem and project state via evaluate_matlab_code, specifically through calls like save_system and slbuild.
Sanitization: The skill uses the introspect_api script to discover valid options, reducing the risk of interpolating arbitrary or malicious strings into MATLAB commands.
[DYNAMIC_EXECUTION]: The skill constructs and executes MATLAB code snippets at runtime using the evaluate_matlab_code tool to interact with the coder.mapping and coder.dictionary APIs. While these follow established vendor patterns for Simulink automation, the assembly of executable strings from project data represents a dynamic execution vector. The skill also includes a pre-compiled MATLAB binary file (scripts/introspect_api.p), which is an opaque artifact that cannot be statically audited.

simulink-configure-code-interfaces