mechanistic-interpretability

mechanistic-interpretability is a plugin for Claude Code from Orchestra-Research/AI-Research-SKILLs. Its manifest loads nothing; the 4 skills it bundles cost 215 tokens per session together, scanned A, original, MIT.

A collection of tools for examining how neural networks and language models work internally. It includes TransformerLens, SAELens, NNSight, and pyvene.

In plain words
What is it for?
Use it to analyze model internals, study how models compute, and look for circuits or other recurring internal patterns.
Why use it?
It helps you investigate a model’s internal computations instead of treating its outputs as a black box.

Plugin for Claude Code

Written for Claude Code: a Claude Code plugin manifest.

Good fit Use it to analyze model internals, study how models compute, and look for circuits or other recurring internal patterns.

Compare 6 plugins from other repositories ↓
Install in Claude Code
/plugin marketplace add Orchestra-Research/AI-Research-SKILLs/plugin install mechanistic-interpretability
About the project

AI Research Skills Library is a collection of reusable instructions that guide AI agents through research and machine-learning engineering tasks, from finding ideas and writing papers to training, evaluation, and deployment. It is for configuring agents such as Claude Code, Codex, and Gemini to perform research workflows.

Orchestra-Research/AI-Research-SKILLs · 12,412 stars · on GitHub · orchestra-research.com

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Clone the repo
git clone --depth 1 https://github.com/Orchestra-Research/AI-Research-SKILLs

Made for: Claude Code.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for mechanistic-interpretability

README.md
[![agentmods](https://agentmods.dev/badge/plugins/orchestra-research/ai-research-skills/mechanistic-interpretability/github.svg)](https://agentmods.dev/plugins/orchestra-research/ai-research-skills/mechanistic-interpretability)
Your own site
<a href="https://agentmods.dev/plugins/orchestra-research/ai-research-skills/mechanistic-interpretability"><img src="https://agentmods.dev/badge/plugins/orchestra-research/ai-research-skills/mechanistic-interpretability/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for mechanistic-interpretability

Your own site · 80×15
<a href="https://agentmods.dev/plugins/orchestra-research/ai-research-skills/mechanistic-interpretability"><img src="https://agentmods.dev/badge/plugins/orchestra-research/ai-research-skills/mechanistic-interpretability.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session not measured What this adds to a session before it is invoked.
When invoked not measured The manifest loads nothing itself; its 4 skills cost 215 tokens a session between them.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Security

Grade A, and why

mechanistic-interpretability scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.claude-plugin/marketplace.json#mechanistic-interpretability · 12 lines

What it actually says

{
  "name": "mechanistic-interpretability",
  "description": "Neural network interpretability tools including TransformerLens, SAELens, NNSight, and pyvene. Use when analyzing model internals, finding circuits, or understanding how models compute.",
  "source": "./",
  "strict": false,
  "skills": [
    "./04-mechanistic-interpretability/nnsight",
    "./04-mechanistic-interpretability/pyvene",
    "./04-mechanistic-interpretability/saelens",
    "./04-mechanistic-interpretability/transformer-lens"
  ]
}
Contents

What it installs

The manifest is a name and a version. 4 skills travel with it, and installing the plugin installs all of them — 215 tokens a session between them. Each is measured on its own page, and each can be installed alone.

Files

What ships with it

1 file beside marketplace.json#mechanistic-interpretability in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 5d ago First seen · 12 lines scan A bf1a5a4557c9

Subscribe to this mod's changes

mechanistic-interpretability is a plugin published in the GitHub repository Orchestra-Research/AI-Research-SKILLs (12,412 stars, last pushed 2mo ago), licensed MIT. Its token cost is not measured: this kind of file is read by the harness, not the model. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.

Related

Other plugins, from other repositories

medsci-modeling

Clinical AI model research engineering: choose a paper-grounded architecture, scaffold a reproducible PyTorch training repo, validate the model's split and validation design, document it (Model Card / Datasheet), and evaluate models or LLMs/MLLMs on clinical tasks. Integrates MONAI / nnU-Net, never replaces them.

Aperivue/medsci-skills · not measured

medsci-data

Study design and validity review, literature-grounded variable operationalization, sample-size planning, data cleaning, de-identification, codebook generation, and dataset versioning.

Aperivue/medsci-skills · not measured

geoai-skills

GeoAI-powered skills for Claude Code: inspect geospatial files, download satellite imagery, search STAC catalogs, fetch Overture Maps data, process rasters, run AI object detection, and search session logs.

opengeos/geoai-skills · not measured

tml

tuneml — a read-first advisor for training and evaluating neural networks well: the scientific method for tuning, and the speed↔quality frontier, in one place. Prices every candidate change in time-to-target-quality rather than throughput, grades the evidence behind it, and names what it would silently break. Five…

emaballarin/ccplugins · not measured

r-tidy-modeling

Expert agents for R data science and biostatistics using tidyverse and tidymodels.

choxos/BiostatAgent · not measured

tidymodels-developers

Development tools and guides for tidymodels ecosystem packages including yardstick metrics, recipes steps, dials parameters, and parsnip models.

tidymodels/skills · not measured