The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/run_finetuning.py script uses subprocess.run to invoke the mattergen-finetune command-line utility. This is used to initiate the model training process and is a standard integration pattern for wrapping CLI-based machine learning tools.
[SAFE]: The skill dynamically generates Hydra configuration files within the mattergen library's configuration path to register custom properties for fine-tuning. While this involves modifying files within the environment, it is the documented method for extending MatterGen's conditioning capabilities and is performed in a controlled manner based on user-provided training parameters.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface where user-supplied property names and training data are used to generate configuration files. 1. Ingestion points: scripts/prepare_training_data.py (input structures) and scripts/run_finetuning.py (property name argument). 2. Boundary markers: None present; user input is interpolated directly into YAML templates. 3. Capability inventory: subprocess.run in scripts/run_finetuning.py executes system commands. 4. Sanitization: None observed for the property_name parameter, though this is typical for scientific orchestration scripts designed for local execution by researchers.

ml-generative-mattergen