The Agent Skills Directory

[SAFE]: No security issues or malicious patterns were detected. The skill instructions and associated scripts define a legitimate scientific research workflow.
[EXTERNAL_DOWNLOADS]: The skill and its examples refer to academic datasets and research tool documentation hosted on well-known platforms such as GitHub and GitLab (e.g., github.com/Huamh/Data-for-cluster-expansion-of-PdPtAg-ternary-alloy and gitlab.com/materials-modeling/icet). These are treated as neutral references to community-standard scientific resources.
[COMMAND_EXECUTION]: The skill uses Python scripts (e.g., prepare_disordered.py, extract_mc_structures.py) to process structural data. These scripts utilize standard libraries for scientific computing, such as pymatgen, smol, and numpy, to perform geometric and chemical analysis of crystal structures.
[INDIRECT_PROMPT_INJECTION]: The skill processes external scientific data files (CIF, JSON, HDF5, and ASE database files).
Ingestion points: Data enters the agent's context through structural input files like input_structure.cif and training_data.json during model training and MC trajectory extraction.
Boundary markers: The workflow relies on data being parsed by specialized scientific libraries rather than natural language instructions, reducing the risk of accidental obedience to embedded instructions.
Capability inventory: Agent capabilities are scoped to scientific simulation tasks (fitting models, running Monte Carlo steps) via dedicated MCP tools.
Sanitization: Input data is validated and structured through the use of established structural parsers (Pymatgen, ASE), which enforce strict schemas for atomic coordinates and species information.

ml-cluster-expansion