The Agent Skills Directory

[SAFE]: The skill implements a 'plan-but-do-not-execute' workflow. It provides Python-based helper scripts such as scripts/plan_batch_command.py that generate command-line arguments for MATLAB/Octave but explicitly forbid their automatic execution by the agent to maintain a human-in-the-loop safety boundary.
[SAFE]: Robust input validation is implemented in scripts/_common.py, which enforces strict limits on input sizes, rejects symlinks to prevent path traversal attacks, and restricts all operations to the local filesystem by rejecting URLs in path strings.
[SAFE]: The skill includes a dedicated security scanner (scripts/scan_m_code.py) designed to detect dangerous MATLAB functions like eval, system, and mex entry points, along with opaque artifacts such as .mlx and MEX binaries before they are processed.
[SAFE]: Data exchange is handled via a specialized inventory tool (scripts/inventory_mat_file.py) that inspects MAT file metadata using scipy or h5py backends without triggering potentially malicious object deserialization or code execution during the triage phase.
[SAFE]: The skill correctly handles external dependencies by referencing well-known scientific libraries (scipy, h5py) and the official matlabengine package, avoiding untrusted or obfuscated sources.

matlab