Menu

Individual skill

mechanistic-interpretability

0upvotes

Reverse-engineer how language models think — features, circuits, and neurons — to explain and audit their behavior.

skills/mechanistic-interpretability/SKILL.md

Read pinned source

Source commit faf37c0859ebf627bfcb5225e8057d088691e0ae

What was checked

Matches the observed repository HEAD
Yes
Scanner rules are current
No
Repository scan complete
No
Skill scan complete
Yes
Observed behavior
Not checked yet
Last checked
Nov 8, 2026

These are observations at the recorded commit, not a guarantee of safety. Review the source before running it.

Install this skill

Choose an agent in your Stack or copy a command below. Commands fetch the pinned commit and copy only this skill directory. Existing destinations are refused. Copying a command does not install anything.

An install command is unavailable: current rules, complete repository and skill scans, an eligible behavior level and a permitted license are required.

Repository context

Repository stars (not skill ratings)
56
Repository last updated
Oct 10, 2026
License
MIT

Observed agent integration: No specific integration evidence. Standard SKILL.md installation does not imply verified compatibility.

Files and excerpt

File metadata and permitted, redacted excerpts come from the recorded scan. Source links are pinned to that commit.

Browse files
Read excerpt

## Overview

Mechanistic interpretability treats a neural network like a program to be decompiled
rather than a black box to be prodded. Instead of asking "what does the model output,"
it asks "which internal structures cause this output." The core insight is that
networks learn reusable, interpretable features — directions in activation space
that correspond to concepts — and compose them into circuits: small subgraphs of
neurons and attention heads that implement recognizable algorithms.

Progress in the field has moved from single neurons (often polysemantic, doing
several jobs at once) to features extracted with sparse autoencoders, which
decompose activations into a larger, cleaner dictionary of monosemantic directions.
For transformers, canonical circuits include induction heads (copying recent
patterns), attention heads that move subject tokens for factual recall, and refusal
Back to top