The Agent Skills Directory

[PROMPT_INJECTION] (LOW): Potential for indirect prompt injection via untrusted scientific data files.
Ingestion points: The scripts structure_analyzer.py and structure_converter.py use Structure.from_file() to ingest data from external files.
Boundary markers: No boundary markers or 'ignore' instructions are present to delimit data from instructions within the files.
Capability inventory: The skill is limited to scientific calculations, symmetry analysis, and file format conversion; it lacks high-risk capabilities like arbitrary command execution or external network exfiltration.
Sanitization: No sanitization is performed on internal file comments or metadata that might be processed by a downstream agent.
[CREDENTIALS_UNSAFE] (SAFE): The phase_diagram_generator.py script correctly retrieves the Materials Project API key from environment variables using os.environ.get('MP_API_KEY').

pymatgen