The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, including Touchstone S-parameter files (e.g., .s4p, .s12p), CTLE frequency response data in CSV format, and captured waveforms from external sources like oscilloscopes.
Ingestion points: Data enters the environment via functions like SParameterChannel, ctlefit.readcsv, and waveform processing scripts.
Capability inventory: The skill uses evaluate_matlab_code and run_matlab_file to perform complex calculations, file writes, and system commands.
Sanitization: There are no specific instructions or boundary markers provided to sanitize or validate the content of these external files against embedded malicious instructions.
Boundary markers: While the instructions are technical, they lack explicit warnings to the agent to disregard natural language instructions that might be embedded in metadata or comments within these data files.
[COMMAND_EXECUTION]: The skill provides instructions for programmatic build processes that involve system-level commands, such as using slbuild and serdes.AMIExport to compile C++ source code into shared libraries (.dll, .so).
Evidence: The documentation in reference/programmatic-ami-generation.md includes a pattern to clear the NoDefaultCurrentDirectoryInExePath environment variable to allow the build system to find local batch files (Tx.bat), which is a legitimate requirement for the toolbox's compilation workflow in restricted environments.

matlab-model-serdes-systems