The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing the torch-geometric library and its optimized extension wheels (pyg-lib, torch-scatter, etc.) from the official wheel index at data.pyg.org. It also demonstrates the use of built-in datasets (Planetoid, TUDataset, ShapeNet) which download graph data from standard research repositories.
[INDIRECT_PROMPT_INJECTION]: The documentation describes how to ingest external data from CSV files and remote URLs via the Dataset and InMemoryDataset classes. This introduces a data ingestion surface, though the skill correctly advises users to use trusted sources and verify data integrity before loading.
Ingestion points: references/custom_datasets.md (loading CSV via pandas, download_url usage).
Boundary markers: Absent in code snippets, but instructional text advises source verification.
Capability inventory: File reads (CSV), network downloads (download_url).
Sanitization: Relies on standard pandas and PyTorch loading mechanisms.
[DYNAMIC_EXECUTION]: The SequenceEncoder example in references/custom_datasets.md utilizes the SentenceTransformer library to dynamically download and load pre-trained embedding models from Hugging Face. This is standard functionality for machine learning workflows.

torch-geometric