neurotrace

Automate end-to-end causal tracing workflows to locate knowledge neurons in encoder-only transformers.

2|Updated Feb 9, 2026
One-click install
npx skills add https://github.com/cristianleoo/NeuroTrace --skill neurotrace
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: neurotrace
Source: https://github.com/cristianleoo/NeuroTrace/tree/main
Command: npx skills add https://github.com/cristianleoo/NeuroTrace --skill neurotrace

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Automates the location of factual knowledge in encoder-only transformers by orchestrating end-to-end causal tracing workflows.

Core Features & Use Cases

  • Orchestrates dataset building, tracing, charting, and interpretation for encoder-based models.
  • Enables model comparisons, domain-specific knowledge testing, and generation of publication-ready charts.
  • Supports both manual and agent-driven workflows via a CLI or Strands orchestrator.

Quick Start

Run an end-to-end NeuroTrace workflow to identify knowledge locations across models and generate trace charts in one go.

Frequently Asked Questions about neurotrace

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I locate factual knowledge in encoder-only transformers?

Causal tracing locates factual knowledge in encoder-only transformers by orchestrating dataset building, tracing, and charting to pinpoint knowledge neurons across models.

What is activation patching for encoder models?

Activation patching for encoder models is a causal tracing technique used to identify knowledge neurons, automating the process of locating where factual information is stored.

Can I generate publication-ready charts for transformer model comparisons?

Yes, you can generate publication-ready trace charts for transformer model comparisons by running end-to-end workflows that build datasets, trace activations, and visualize results.

Does causal tracing work with agent-driven experiments across datasets?

Yes, causal tracing supports agent-driven experiments across datasets via a Strands orchestrator or manual CLI, enabling model comparisons and domain-specific knowledge testing.

What is the best way to automate end-to-end causal tracing workflows?

The best way to automate end-to-end causal tracing workflows is using a built-in CLI or Strands orchestrator, which handles dataset generation, tracing, charting, and interpretation in one go.

Do I need dependencies to run causal tracing on encoder-only transformers?

No dependencies are required to run causal tracing on encoder-only transformers, as the workflow automates dataset generation, tracing, and visualization natively through its CLI or orchestrator.