The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses importlib and exec_module to load and run arbitrary Python code from generator files (e.g., .step.py). This is found in scripts/packages/cadgen/src/cadgen/_internal/generation_runner.py.
[DYNAMIC_EXECUTION]: In scripts/packages/implicitjs/src/lib/implicitCad/loader.js, the skill uses dynamic import() on data URLs created from source strings, and in scripts/packages/implicitjs/src/lib/implicitCad/sdfEvaluator.js, it uses new Function to compile and execute logic from GLSL strings.
[COMMAND_EXECUTION]: The function _rebuild_child_in_subprocess in scripts/packages/cadgen/src/cadgen/_internal/generation.py executes shell commands using subprocess.run with a dynamically generated Python bootstrap string.
[OBFUSCATION]: Several JavaScript files, such as scripts/packages/cadjs/bin/implicit-artifact.mjs and scripts/packages/cadjs/bin/meshWorkerEntry.js, contain very large minified or packed data strings and custom decoding logic, which obscures the underlying executable content (likely WebAssembly modules).
[INDIRECT_PROMPT_INJECTION]: The skill's architecture involves processing untrusted CAD data and executing code from user-controlled files. This creates a significant surface for indirect prompt injection where malicious instructions could be embedded in data files.
Ingestion points: Processes .implicit.js, .step, and .dxf files provided in the workspace.
Boundary markers: Absent; the skill processes logic within these files directly.
Capability inventory: Includes arbitrary Python execution via exec_module, JavaScript execution via new Function, and shell command execution via subprocess.
Sanitization: None; the skill is designed to faithfully execute the provided modeling logic.

implicit-cad