The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interpolates user-provided values such as function names, server addresses, and file paths directly into shell-executed commands, including matlab -batch and curl.
Evidence: In Step 5, the <fcnName>, <scheme>, <host>, <port>, and <absolute-folder-path> variables are interpolated into a MATLAB command string passed to the -batch flag.
Evidence: Step 4.5 uses curl with interpolated <scheme>, <host>, <port>, and <archiveName> to perform a JSON-RPC ping to the user-specified server.
[DYNAMIC_EXECUTION]: The skill generates MATLAB code at runtime, sometimes writing it to temporary .m files, and then executes it using matlab -batch.
Evidence: Step 5 explicitly instructs the agent to create a temporary script for complex builds and run it with matlab -batch "run('<path>')".
[PERSISTENCE]: The skill modifies the agent's global configuration file to ensure the newly created tools are automatically loaded in future sessions.
Evidence: Step 7 describes the process for reading and writing to ~/.claude.json to add new entries under the mcpServers key.
[DATA_EXFILTRATION]: The skill accesses the agent's sensitive configuration files, which may contain information about other configured tools or environment settings.
Evidence: Step 7 requires the agent to read the contents of ~/.claude.json to register the new MCP server.
[INDIRECT_PROMPT_INJECTION]: The skill processes external .m files and user-defined resource strings, creating a potential surface for indirect injection attacks.
Ingestion points: User-provided .m function files (Step 2/3), server resources (Step 2), and build manifests (Step 4.1).
Boundary markers: The instructions do not specify any delimiters or safety warnings for the agent when processing the content of these external inputs.
Capability inventory: The skill utilizes shell execution (matlab, curl, rm, cp, mkdir) and the ability to modify the agent's ~/.claude.json configuration file.
Sanitization: While the skill performs path existence and filename case validation, it lacks explicit sanitization for the content of the functions or resource strings before they are incorporated into the build and deployment process.

mps-mcp-build