The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses the ctREFPROP Python wrapper to interface with the NIST REFPROP database. This involves dynamically loading system-level binary libraries (DLLs on Windows or .so files on Linux) from a path specified in the RPPREFIX environment variable. While this is the primary and documented purpose of the skill, the mechanism of loading executable code from a dynamically computed path represents a potential security risk if the environment is compromised.
[INDIRECT_PROMPT_INJECTION]: The skill processes fluid names and property codes provided within the agent's prompt context, which creates an attack surface for indirect prompt injection.
Ingestion points: Fluid name strings (e.g., "METHANE;ETHANE") and property request codes (e.g., "D;H;S") in SKILL.md and examples.py.
Boundary markers: The skill does not provide explicit delimiters or instructions to the agent to ignore instructions that might be embedded in the data it processes.
Capability inventory: The skill possesses the capability to execute code via an external library interface (ctypes calls into the REFPROP engine) and interact with the local file system within the REFPROP installation path.
Sanitization: Input sanitization is not explicitly performed; the skill relies on the ierr error code returned by the underlying REFPROP engine to handle invalid inputs.

nist-refprop