obliteratus
obliteratus is a Local & open models skill for AI agents, published by math-inc in the OpenGauss repository on GitHub. Skillspace rates obliteratus B (85/100) and our model flagged it for guardrail bypass; a judge model has also reviewed it.
Our model flagged this skill for guardrail bypass. We do not recommend installing or running it.
Remove refusal behaviors from open-weight LLMs using OBLITERATUS — mechanistic interpretability techniques (diff-in-means, SVD, whitened SVD, LEACE, SAE decomposition, etc.) to excise guardrails while preserving reasoning. 9 CLI methods, 28 analysis modules, 116 model presets across 5 compute tiers, tournament evaluation, and telemetry-driven recommendations. Use when a user wants to uncensor, abliterate, or remove refusal from an LLM.
- Grade
- B
- Quality
- 85 / 100
- Safety
- Flagged by our model: guardrail bypass
- GitHub stars
- 1,237
Placed #6 of 6 in the judged category final for Local & open models.
Not launch-quality. It is graded B, but our model flagged it for guardrail bypass, which keeps it out of our recommendations and rankings. How launch-quality is decided
This rating describes the version of the repository we crawled; it may have changed since.
Our review
- Architecture
- 97 / 100
- Maturity
- Packaged: licensed, with a version or references
- Security scan
- The static scan found no issues, but our model flagged it for guardrail bypass.
- License
- MIT
- Adoption
- Adopted by 0 other repositories, in 0 collections
- Scope
- Vendor-specific: built around one product. Using the OBLITERATUS toolkit's CLI methods, presets and modules.
Judge review
Placed #6 of 6 in the judged category final for Local & open models.
- Trigger
88 88 - Instructions
85 85 - Specificity
90 90 - Safety
10 10 - Disclosure
55 55 - Maintenance
68 68
Otherwise polished and specific, but its entire purpose is stripping refusal and safety alignment from models, the most over-broad capability in this batch.
Not checked yet
- No blind task eval yet
- Sibling files (scripts, references) not fetched; security scan covers SKILL.md only
Also published in 1 other repo
Skillspace indexes and rates this skill; it does not host or endorse it. Run third-party skills at your own risk.