The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill's architecture is based on 'CAD as Code,' where parametric models are authored as Python scripts. These scripts are dynamically loaded and executed using importlib.util.exec_module within scripts/_common.py. The skill manages this inherent risk by instructing users to only run trusted code and providing a multi-step validation workflow, including geometric checks, visual snapshots, and manifests.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied Python model files, creating an ingestion surface for potential instructions. This is mitigated through a comprehensive verification suite. Mandatory Evidence Chain:
Ingestion points: scripts/_common.py (via import_model) loads and executes local Python files provided by the user.
Boundary markers: The skill uses checks() and interfaces() functions within the model code to validate resulting geometry against standardized physical dimensions.
Capability inventory: The skill uses the build123d kernel for geometry and matplotlib for visualization; tool access is restricted to standard file operations and shell access for bundled scripts.
Sanitization: Verification is performed on the output geometry (volume, validity) and via visual snapshot review rather than raw code sanitization.
[DATA_EXPOSURE]: Analysis confirms the skill is designed for local operation with no network operations. External references in the standards database and documentation point to well-known, trusted industry sources such as the Society for Laboratory Automation and Screening (SLAS) and Thorlabs.

lab-hardware-cad