The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through its data ingestion and processing workflow.
Ingestion points: The skill is instructed in SKILL.md to analyze input data via the Model Context Protocol (MCP) to determine spectrum, signal length, and interference locations.
Boundary markers: The instructions define a structured workflow (Phases 1-5) to guide the agent, but do not specify the use of delimiters or 'ignore embedded instructions' warnings when processing external signal metadata or related data strings.
Capability inventory: The manifest.yaml grants the skill powerful execution capabilities including evaluate_matlab_code, check_matlab_code, and run_matlab_file, which allow for the execution of arbitrary MATLAB scripts.
Sanitization: While the workflow focuses on numerical and spectral analysis, which naturally limits typical text-based injection, the agent lacks explicit instructions to sanitize or validate non-numerical metadata that might be processed alongside the signal data.
[SAFE]: The skill demonstrates high-quality engineering by enforcing numerical stability (SOS form for high-order IIR filters) and computational efficiency (recommending fftfilt for long FIR filters and multirate/IFIR approaches for narrow transitions).
[SAFE]: All external references, licenses, and documentation links point to official MathWorks domains, consistent with the vendor identity.

matlab-design-digital-filter