transformer-lens-interpretability
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides instructions to install the
transformer-lenslibrary from its official sources, including the standard Python Package Index (PyPI) and the official GitHub repository for the project. These are established and recognized resources within the AI research community. - [INDIRECT_PROMPT_INJECTION]: The skill describes workflows that involve passing natural language strings to transformer models to analyze internal activations (e.g., induction head detection and activation patching).
- Ingestion points: Prompts are provided in code examples within
SKILL.mdandreferences/tutorials.mdas input to tokenization functions. - Boundary markers: None; strings are directly tokenized for numerical analysis.
- Capability inventory: The skill's functionality is restricted to mathematical analysis and visualization using PyTorch; it contains no instructions for arbitrary shell execution, unauthorized network requests, or sensitive file system modifications.
- Sanitization: None; input strings are converted to token IDs for model processing, which is standard for interpretability research.
Audit Metadata