The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/test_digest_tools.py utilizes subprocess.run to invoke the Houdini Python interpreter (hython). This execution is conditional on the HYTHON environment variable being set.
[DYNAMIC_EXECUTION]: The reference file reference/python_sop_traps.md describes the use of exec() for running helper scripts within Houdini Python SOPs. It correctly identifies this as a 'trap' due to potential namespace clobbering and recommends using private namespaces or function wrappers as mitigation.
[DYNAMIC_EXECUTION]: The test suite in scripts/test_digest_tools.py dynamically generates a Python script (defined in the SCENE variable), writes it to a temporary file, and executes it via a subprocess call.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and analyze external Houdini scene files (.hip) and JSON-formatted geometry digests, which presents a surface for processing untrusted data.
Ingestion points: geo_digest.py and hip_provenance.py ingest file paths and data from the Houdini environment.
Boundary markers: None are explicitly implemented to separate data from instructions during processing.
Capability inventory: The skill possesses file system read/write capabilities (via hou and os modules) and command execution capabilities (within the test scripts).
Sanitization: The implementation uses cryptographic hashing (MD5, SHA256) and fixed-precision rounding to validate the identity and integrity of geometry data.

houdini-python-automation