The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of well-known and widely used machine learning packages from official registries (PyPI), including lightning, wandb, mlflow, comet-ml, and deepspeed. These references are standard for the intended use case of neural network training.
[INDIRECT_PROMPT_INJECTION]: The skill provides LightningDataModule templates designed to ingest external data for machine learning tasks. This represents a potential vulnerability surface where malicious instructions could be embedded in datasets and executed by an agent.
Ingestion points: Data enters the system through the prepare_data and setup methods defined in references/data_module.md and scripts/template_datamodule.py.
Boundary markers: The provided boilerplate code does not include specific boundary markers or instructions to the agent to disregard natural language commands found within the data.
Capability inventory: The skill possesses Bash, Write, Read, and Edit capabilities, which could be exploited if the agent is influenced by malicious data content to perform unauthorized file or system operations.
Sanitization: The templates lack explicit input validation or sanitization logic to filter out potentially harmful content from the ingested datasets.

pytorch-lightning