The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to download pre-compiled binaries and source code from the official MATLAB GitHub repository. These sources are associated with a well-known technology vendor (MathWorks) and are documented neutrally.
Evidence: https://github.com/matlab/matlab-mcp-server/releases/latest/download/matlab-mcp-server-macos-amd64, https://github.com/matlab/matlab-mcp-server/releases/latest/download/matlab-mcp-server-macos-x64, and github.com/matlab/matlab-mcp-server/cmd/matlab-mcp-server@latest.
[COMMAND_EXECUTION]: The skill enables an AI agent to execute arbitrary code within the MATLAB environment on the host machine. This grants the agent full access to MATLAB's capabilities, including file system operations and network requests, within the scope of the user's permissions.
Evidence: The execute_matlab_code and evaluate_matlab_expression tools allow the agent to run any provided string as MATLAB code.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests external code for execution without visible sanitization or structural boundaries. If the agent incorporates untrusted data from other sources into its MATLAB commands, it could lead to the execution of unintended or malicious code.
Ingestion points: execute_matlab_code ('code' argument) and evaluate_matlab_expression ('expression' argument).
Boundary markers: Absent in the provided documentation.
Capability inventory: MATLAB execution allows file reads/writes, system command execution (via !), and network operations.
Sanitization: No sanitization or validation of the input strings is described in the skill metadata.
[DYNAMIC_EXECUTION]: The server acts as a runtime environment for dynamic code generation and execution, which is the primary intended functionality of the skill.

matlab-mcp-server