Transformer Lens Interpretability

If you're doing mechanistic interpretability work on transformer models, this is the standard library you'll reach for. TransformerLens gives you clean hooks into every activation layer so you can do activation patching, study attention patterns, cache intermediates, and reverse-engineer learned circuits like induction heads. It's Neel Nanda's library with over 2,900 stars, now maintained by Bryce Meyer. The skill wraps it for Claude Code, which is helpful if you're running experiments and want the agent to understand the library's patterns for causal tracing and circuit analysis. It's specialized tooling, so you'll know if you need it.

Install to Claude Code

npx -y skills add davila7/claude-code-templates --skill transformer-lens-interpretability --agent claude-code