What is mechinterp-replication CLAUDE.md?

A framework for repeating mechanistic interpretability research across different open-source AI model families and sizes. Mechanistic interpretability studies the internal processes that produce a model's outputs.

Use it to configure, run, and document replication experiments across models such as Llama, Qwen, and Gemma.

It helps test whether findings from one research paper also hold for other models instead of assuming they generalize.

How zachgoldfine44/mechinterp-replication configures Claude Code

Reuse

Borrowing it

Nothing to install: this file belongs to zachgoldfine44/mechinterp-replication. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file

curl -O https://raw.githubusercontent.com/zachgoldfine44/mechinterp-replication/main/CLAUDE.md

Clone the repo

git clone --depth 1 https://github.com/zachgoldfine44/mechinterp-replication

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.

README.md

[![agentmods](https://agentmods.dev/badge/instructions/zachgoldfine44/mechinterp-replication/claude-md/github.svg)](https://agentmods.dev/instructions/zachgoldfine44/mechinterp-replication/claude-md)

Your own site

<a href="https://agentmods.dev/instructions/zachgoldfine44/mechinterp-replication/claude-md"><img src="https://agentmods.dev/badge/instructions/zachgoldfine44/mechinterp-replication/claude-md/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.

Your own site · 80×15

<a href="https://agentmods.dev/instructions/zachgoldfine44/mechinterp-replication/claude-md"><img src="https://agentmods.dev/badge/instructions/zachgoldfine44/mechinterp-replication/claude-md.svg" alt="Reviewed on agentmods" width="80" height="20"></a>

Per session 9,216 This file is loaded in full into every session.

When invoked 9,216 The same file — it is already loaded in full.

Security scan A 1 finding. A grade says what 26 rules found in the file — not that it is safe.

Origin original No closer match found in the catalogue.

Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

Model	Per session	Once invoked
Fable 5.1	$0.09216	$0.09216
Opus 5.5	$0.03686	$0.03686
Sonnet 5.5	$0.01843	$0.01843
Haiku 4.5	$0.00922	$0.00922

Measured 2mo ago against content hash 04137e6b1fa1, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

mechinterp-replication CLAUDE.md scanned grade A with 1 finding 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 2mo 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.

Runs shell commandslowCapability

Expected in a hook, worth knowing in a rule or an instructions file.

subprocess.run(["git", "clone",

CLAUDE.md · 769 lines

How it starts

The opening of the file, as written. The whole thing — 769 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Mechinterp Replication Engine — Operational Guide

For humans: If you're looking for an overview of what this project is and how to use it, see README.md. If you want to contribute a replication or improve the harness, see CONTRIBUTING.md.

This file is the operational guide for Claude Code agents (and power-users who want to understand the internal conventions). It documents session workflow, commit cadence, storage layout, and coding standards that the autonomous development pipeline relies on.

What is this?

A general-purpose framework for replicating mechanistic interpretability papers across multiple open-source model families (Llama, Qwen, Gemma) at various scales, then writing up the results.

Think of it like a replication study in psychology: same experimental protocol, different populations (models), measuring what transfers and what doesn't. But instead of one paper hardwired in, this is a replication engine — you feed it a paper config, and it designs experiments, runs them, and writes up whether the findings generalize.

Quick reference

Design document: DESIGN.md (framework architecture, replication protocol)
Mech-interp guardrails: GOTCHAS.md (preflight + post-result checklists — read before designing experiments and after every positive result)
Harness changelog: CHANGELOG.md (framework-level changes only — src/, tests/, scripts/, repo-level docs)
Per-replication milestone log: writeup/{paper_id}/{replication_id}/PROGRESS.md (big-picture state of THIS specific replication — read first every session, after you know which replication you're on)
Per-replication changelog: writeup/{paper_id}/{replication_id}/CHANGELOG.md (append-only work-unit log for this replication — your incremental memory)
Active paper: config/papers/{paper_id}/ — paper text + stimuli, with one subfolder per replication under replications/
Active paper config: config/papers/{paper_id}/replications/{replication_id}/paper_config.yaml (per-replication; paper-level paper_config.yaml exists in legacy layouts)
Paper as ground truth: config/papers/{paper_id}/paper.md (full paper text — see "Paper as oracle" below)
Model matrix: config/models.yaml

Read the full file on GitHub · 769 lines

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.

2mo ago First seen · 769 lines · 9,216 tokens per session scan A 04137e6b1fa1

Subscribe to this mod's changes

mechinterp-replication CLAUDE.md is an instructions file published in the GitHub repository zachgoldfine44/mechinterp-replication (13 stars, last pushed 6mo ago), licensed MIT. It adds 9,216 tokens to every session, about $0.0369 per session on Opus 5.5. A static security scan graded it A with 1 finding (runs shell commands). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Data and AI Scientific

mechinterp-replication CLAUDE.md next to 6 instructions from other repositories
Mod	Type	Stars	Last push	Licence	Scan	Tokens	Works with
mechinterp-replication CLAUDE.mdzachgoldfine44/mechinterp-replication	Instructions file	13	6mo ago	MIT	A	9,216	Claude Code
vscode buildNext.instructions.mdmicrosoft/vscode	Instructions file	195k	yesterday	MIT	A	9,719	GitHub Copilot
next.js AGENTS.mdvercel/next.js	Instructions file	143k	yesterday	MIT	A	7,461	Codex, OpenCode
langchain AGENTS.mdlangchain-ai/langchain	Instructions file	149k	yesterday	MIT	A	4,680	Codex, OpenCode
vscode oss-third-party-notices.instructions.mdmicrosoft/vscode	Instructions file	195k	yesterday	MIT	A	5,516	GitHub Copilot
vscode copilot-instructions.mdmicrosoft/vscode	Instructions file	195k	yesterday	MIT	A	3,093	GitHub Copilot
deepseek-harness AGENTS.mddeepseek-ai/deepseek-harness	Instructions file	274k	2d ago	MIT	A	4,313	Codex, OpenCode

mechinterp-replication: Instructions file for Claude Code