The Agent Skills Directory

[SAFE]: The skill provides legitimate guidance and documentation for mechanistic interpretability research using the TransformerLens library. All analyzed components align with standard research and development practices.- [EXTERNAL_DOWNLOADS]: The skill provides instructions for installing the transformer-lens library via pip from PyPI and the official GitHub repository (TransformerLensOrg/TransformerLens). These are well-known and reputable sources for this research tool.- [CREDENTIALS_UNSAFE]: The documentation includes the use of os.environ["HF_TOKEN"] = "your_token". This uses a clear placeholder string and is standard practice for authenticating with Hugging Face to access gated models.- [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves processing text prompts and model activations for analysis. While this represents a data ingestion surface, it is a core requirement of the library's research purpose and does not pose an atypical security risk.- [DYNAMIC_EXECUTION]: The skill utilizes HookPoints to execute Python functions during model inference. The provided examples (ablation, patching, and steering) demonstrate legitimate and safe use of this feature for research analysis.

transformer-lens-interpretability