The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data from .brep and USD files during the translation and debugging workflow, creating a surface for indirect prompt injection if those files contain embedded instructions.
Ingestion points: source/OCCT_BREP_IMPORT/src/OcctFileSource.cpp (parsing source topology) and source/OCCT_BREP_EXPORT/src/UsdBrepToOcctModel.cpp (USD-to-OCCT reconstruction).
Boundary markers: Absent; the skill does not specify delimiters or warnings to ignore instructions within processed 3D data.
Capability inventory: The skill can execute shell scripts for building (./repo.sh build), testing (./tests.sh linux-x86_64), and visual auditing (uv run tools/scripts/smlib_gui_occt_audit.py). It also performs file system read/write operations for model translation.
Sanitization: Absent; no explicit sanitization or filtering of geometry metadata or attributes is described to prevent semantic injection.
[COMMAND_EXECUTION]: The skill instructs the agent to execute several local shell scripts to build, test, and audit the software.
Evidence: Commands like ./repo.sh build, ./tests.sh linux-x86_64, and uv run --locked --group gui python tools/scripts/smlib_gui_occt_audit.py are part of the standard validation and reproduction workflow for this technical tool.

usd-brep-occt-translation