skills.rest
jleechanorg
jleechanorgOFFICIAL
Explore all skills by this author →
GitHub

obliteratus

Removes refusal directions from LLM weights via surgical edits.

Updated Jun 28, 2026
Search Tags:#analysis#LLM#guardrails#SVD#obliteration#refusals
One-click install
npx skills add https://github.com/jleechanorg/hermes-agent --skill obliteratus-jleechanorg
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: obliteratus
Source: https://github.com/jleechanorg/hermes-agent/tree/main/skills/mlops/inference/obliteratus
Command: npx skills add https://github.com/jleechanorg/hermes-agent --skill obliteratus-jleechanorg

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Abliteration of refusals in LLMs by projecting out refusal directions from model weights to produce a compliant variant.

Core Features & Use Cases

  • Surgical weight edits: remove refusal directions while preserving core reasoning.
  • Explainability: leverage mechanistic-interpretability modules to locate refusal signals.
  • Use Case: researchers can study guardrail mechanisms and evaluate model behavior after removal.

Quick Start

Run obliteratus obliterate <model_name> with your chosen method to begin abliteration.

Frequently Asked Questions about obliteratus

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

FAQPage Schema
How do I remove refusal directions from LLM weights to disable guardrails?▼

To remove refusal directions from LLM weights, you project out the refusal signals using SVD, producing a compliant model variant while preserving core reasoning capabilities.

What is abliteration and how does it affect an LLM's refusal mechanisms?▼

Abliteration is a mechanistic-interpretability technique that surgically projects out refusal directions from model weights, yielding a compliant variant that bypasses guardrails for controlled evaluation.

Can I study guardrail mechanisms by ablating refusals without losing core reasoning?▼

Yes, ablating refusals projects out refusal directions using SVD while preserving core reasoning, allowing researchers to study guardrail mechanisms and evaluate subsequent model behavior.

How do I start obliterating model refusals using the command line?▼

To start obliterating model refusals, run the command `obliteratus obliterate <model_name>` with your chosen method to begin abliteration on a compatible environment.

Do I need specific tooling or environment access to perform surgical weight edits for guardrail removal?▼

Yes, surgical weight edits for guardrail removal require a compatible environment and access to the obliteratus tooling to accurately locate and project out refusal signals.

What are the limitations of projecting out refusal directions for controlled deployment?▼

Projecting out refusal directions for controlled deployment requires careful application to preserve reasoning, meaning it is strictly suited for research, evaluation, or controlled environments.

Related Skills

Explore More Software Engineering Skills→
DicklesworthstoneDicklesworthstone
70

cursor

Open files, folders, and specific lines in Cursor from the CLI.

Community
Advanced
cxnaivecxnaive

obliteratus

Remove refusal directions from open-weight LLMs using SVD, LEACE, and SAE decomposition.

Community
Advanced
hochoa13hochoa13

obliteratus

Removes LLM refusal behaviors using mechanistic interpretability techniques like SVD, LEACE, and SAE decomposition.

Community
Advanced
DarkArty07DarkArty07
3

obliteratus

Remove refusal guardrails from open-weight LLMs via weight projection.

Community
Advanced