The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill facilitates the creation and execution of local MATLAB scripts (.m files) to wrap custom model objects for compatibility with interpretability functions.
The references/custom-model-wrapper.md file outlines a process to generate a <model-variable-name>FcnHandle.m file.
This mechanism includes a mandatory validation step where the generated handle must reproduce the model's known predictions before it is used in further analysis, mitigating risks associated with code generation.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes user-provided models and tabular data.
Untrusted data enters the agent context through variables like Mdl, Xtrain, and Xtest in SKILL.md.
The skill mitigates this by instructing the agent to inventory all inputs and require explicit user confirmation before proceeding with any analysis.
Capabilities are limited to standard analysis functions within the Statistics and Machine Learning Toolbox.
[COMMAND_EXECUTION]: The skill utilizes MATLAB commands to manage its environment and execute analysis.
It uses addpath to incorporate the directory containing generated wrapper scripts into the MATLAB search path.
It leverages evaluate_matlab_code to call statistical functions such as shapley, lime, and permutationImportance.

matlab-interpret-machine-learning-model