# obliteratus-abliteration

This skill runs OBLITERATUS workflows that analyze and alter refusal behavior in LLMs. Use it when testing models from HuggingFace.

- skill · `type=skill` · Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode · 864 installs · ★ 88
- Author: reason-machines
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/reason-machines/trending-skills/tree/HEAD/skills/obliteratus-abliteration/SKILL.md
- Updated: 2026-10-09
- HTML: https://affpapa.org/en/nevibecode/skills/obliteratus-abliteration

## What it does

The skill connects an agent to OBLITERATUS and exposes model analysis, abliteration, and verification workflows. The toolkit uses SVD or PCA to find refusal directions in hidden states, then removes those directions from model weights. It includes a CLI, Python API, local Gradio interface, and Colab notebook.

The CLI supports basic, advanced, and analysis-informed methods. It can sweep intervention strength, benchmark methods, save output to a chosen directory, and run analysis without modifying the model. A LoRA mode provides a reversible form of ablation.

The environment needs Python 3.10+, PyTorch 2.1+ with CUDA or CPU, `transformers`, `accelerate`, and `gradio>=5.29.0`. Gated HuggingFace models also require an account and token. Those models will not load without the token.

## Install

### Claude Code (native platform)

1. **Skills CLI (Vercel)**
```bash
npx skills add reason-machines/trending-skills --skill obliteratus-abliteration -a claude-code
```

2. **Manually: put the files in place**
- File: `~/.claude/skills/obliteratus-abliteration/SKILL.md` — Copy the whole skill folder; for one project use .claude/skills/obliteratus-abliteration/

### Codex

1. **Skills CLI (Vercel)**
```bash
npx skills add reason-machines/trending-skills --skill obliteratus-abliteration -a codex
```

2. **Manually: put the files in place**
- File: `.agents/skills/obliteratus-abliteration/SKILL.md` — For all projects use ~/.agents/skills/

### Cursor

1. **Skills CLI (Vercel)**
```bash
npx skills add reason-machines/trending-skills --skill obliteratus-abliteration -a cursor
```

2. **Manually: put the files in place**
- File: `.cursor/skills/obliteratus-abliteration/SKILL.md` — Cursor also reads .agents/skills/ and .claude/skills/

### GitHub Copilot

1. **GitHub CLI**
```bash
gh skill install reason-machines/trending-skills obliteratus-abliteration
```

2. **Manually: put the files in place**
- File: `.github/skills/obliteratus-abliteration/SKILL.md`

### Gemini CLI

1. **Skills CLI (Vercel)**
```bash
npx skills add reason-machines/trending-skills --skill obliteratus-abliteration -a gemini-cli
```

2. **Platform CLI**
```bash
gemini skills install https://github.com/reason-machines/trending-skills --path skills/obliteratus-abliteration
```

### Windsurf (Devin Desktop)

**Manually: put the files in place**
- File: `.devin/skills/obliteratus-abliteration/SKILL.md` — The legacy .windsurf/skills/ path still works

### OpenCode

**Skills CLI (Vercel)**
```bash
npx skills add reason-machines/trending-skills --skill obliteratus-abliteration -a opencode
```

### Claude.ai

**Claude.ai**
- Zip the folder and upload it in Settings → Capabilities → Skills

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/reason-machines/trending-skills/tree/HEAD/skills/obliteratus-abliteration/SKILL.md

## When to use it
- remove refusal directions from a language model
- compare ablation methods on one model
- inspect refusal signals across model layers
- sweep intervention strength and review the tradeoff
- save an altered model to a local directory

## Example prompts
- Use `obliteratus-abliteration` to prepare an OBLITERATUS run for a HuggingFace model.
- Analyze refusal directions in `meta-llama/Llama-3.1-8B-Instruct` without changing its weights.
- Benchmark `basic`, `advanced`, and `informed` on the same model, then save the outputs.

## Similar
- [skill-creator](https://affpapa.org/en/nevibecode/skills/skill-creator.md) — skill · 414180 installs
- [writing-for-agents](https://affpapa.org/en/nevibecode/skills/writing-for-agents.md) — skill · 466719 installs
- [media-use](https://affpapa.org/en/nevibecode/skills/media-use.md) — skill · 1100000 installs
- [mcp-builder](https://affpapa.org/en/nevibecode/skills/mcp-builder.md) — skill · 130949 installs
- [subagent-driven-development](https://affpapa.org/en/nevibecode/skills/subagent-driven-development.md) — skill · 237746 installs
- [orchestration](https://affpapa.org/en/nevibecode/skills/orchestration.md) — skill · 349800 installs

From the same set: [reason-machines/trending-skills](https://affpapa.org/en/nevibecode/by/reason-machines/trending-skills.md) — 139 cards

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md