The Agent Skills Directory

[COMMAND_EXECUTION]: The skill leverages the evaluate_matlab_code and check_matlab_code tools to execute and validate MATLAB scripts. This is the core functionality required for its stated purpose of performance profiling and optimization.
Evidence: Found in manifest.yaml under required-tools and demonstrated throughout the measurement templates in references/measurement-templates.md.
[DATA_EXFILTRATION]: The workflow includes templates that write performance data to the local file system using the MATLAB save command. This is used to persist baseline measurements for later comparison and does not involve network exfiltration.
Evidence: references/measurement-templates.md contains the command save('baseline.mat', 'baseline').
[PROMPT_INJECTION]: The skill is designed to ingest and process user-provided MATLAB code (untrusted input) for profiling and optimization. While this presents an indirect prompt injection surface, the skill implements a 'Verify Correctness' step that requires the agent to compare the outputs of original and optimized code, serving as a functional safeguard against logic changes.
Ingestion points: User functions are interpolated into the measurement scripts in references/measurement-templates.md.
Boundary markers: Absent.
Capability inventory: Uses evaluate_matlab_code for execution and save/load for file operations.
Sanitization: Includes automated correctness checks using assert and floating-point tolerance comparisons to ensure the behavior of the code remains consistent.

matlab-optimize-performance