obliteratus

Remove refusal behaviors from open-weight LLMs via CLI-driven abliteration methods.

11|Updated May 17, 2026

Search Tags:#cli #llm #mechanistic-interpretability #refusal-removal #abliteration #model-editing

Ready to run

npx skills add https://github.com/StarryCod/cogitum --skill obliteratus-starrycod

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: obliteratus
Source: https://github.com/StarryCod/cogitum/tree/main/cogitum/data/skills/mlops/inference/obliteratus
Command: npx skills add https://github.com/StarryCod/cogitum --skill obliteratus-starrycod

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Abliteration removes refusal behaviors from open-weight LLMs without retraining or fine-tuning, enabling flexible reasoning and experimentation with safeguards in place.

Core Features & Use Cases

CLI-driven abliteration with 9 methods (basic, advanced, aggressive, spectral_cascade, informed, surgical, optimized, inverted, nuclear) to suit different models and accuracy/speed needs.
28 analysis modules and 116 model presets across multiple compute tiers, plus study templates, to support rigorous experimentation and deployment planning.
Supports reproducible workflows via YAML templates and telemetry-enabled configuration, enabling researchers and engineers to safely experiment with guardrail removal in a controlled, auditable way.

Quick Start

Install obliteratus, choose a model and method, and run the CLI 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 behaviors from an open-weight LLM without retraining?▼

Abliteration removes refusal behaviors from open-weight LLMs without retraining or fine-tuning. It analyzes and edits the model's internal mechanisms directly, enabling flexible reasoning and experimentation with safeguards in place.

What CLI methods are available for abliterating LLM refusals?▼

Abliterating LLM refusals supports 9 CLI methods, including basic, advanced, aggressive, spectral_cascade, informed, surgical, optimized, inverted, and nuclear. These methods suit different models and accuracy or speed requirements.

Can I run abliteration workflows on models with limited compute resources?▼

Abliteration workflows support multiple compute tiers and 116 model presets. This allows you to execute refusal removal across various hardware limitations by selecting appropriate presets and methods.

How do I ensure my LLM guardrail removal experiments are reproducible?▼

Reproducible abliteration workflows are ensured via YAML templates and telemetry-enabled configuration. This supports researchers and engineers in safely experimenting with guardrail removal in a controlled, auditable way.

What is the best way to analyze LLM mechanistic interpretability for refusal removal?▼

Analyzing LLM refusal removal is supported by 28 analysis modules and study templates. These facilitate rigorous experimentation and deployment planning when abliterating refusals via mechanistic interpretability techniques.

Do I need fine-tuning datasets to abliterate model refusals?▼

No, you do not need fine-tuning datasets to abliterate model refusals. The process operates by editing the model directly via CLI-driven workflows, bypassing the need for retraining or fine-tuning data.