The Agent Skills Directory

[COMMAND_EXECUTION]: The MCP server implementation in mcp_server.py uses subprocess.run to invoke its internal CLI script scripts/soc_integrate.py. The command execution is performed safely by passing arguments as a list and using the current Python executable, which mitigates shell injection risks.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes Verilog source files and JSON configuration data provided by the user or the environment, which constitutes an indirect prompt injection surface. A malicious actor could attempt to embed instructions within hardware design comments or port metadata.\n
Ingestion points: Verilog files are read and parsed in tools like soc_extract, soc_integrate, and soc_wrap. JSON port-map configurations are ingested by soc_update and the integrate command via the scripts/soc_integrate.py script.\n
Boundary markers: The skill does not implement specific boundary markers or safety instructions to delimit external content when it is interpolated into the agent's context.\n
Capability inventory: The skill possesses the ability to perform filesystem writes (e.g., in soc_integrate and soc_update commands) and execute local scripts via the subprocess module.\n
Sanitization: The skill relies on standard Python libraries (json, csv) for structured data parsing and uses regular expressions for Verilog port extraction.\n- [DYNAMIC_EXECUTION]: The test suite in tests/test_parser.py utilizes the importlib module to dynamically load and execute the soc_integrate.py script for unit testing. While this is restricted to the development and testing environment, it represents a pattern of runtime code loading.

soc-integrate