The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill architecture in scripts/_common.py uses importlib.util.exec_module to dynamically load and execute Python code from user-defined *_model.py files. This behavior is central to the skill's parametric design capability but allows for arbitrary code execution within the environment. Additionally, the skill modifies sys.path to include the model's directory, which could allow for unintended module shadowing or execution of malicious local scripts.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external content from user-supplied model scripts and the assets/standards.json database. While it provides geometric checks, it lacks sanitization or explicit boundary markers for potential instructions embedded in these data sources. Given the agent's broad file-writing and bash capabilities, this represents a vulnerability surface for indirect prompt injection, though the primary danger is assessed as code execution.

lab-hardware-cad