obliteratus

OBLITERATUS: abliterate LLM refusals (diff-in-means).

Quality

67%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Critical

Do not install without reviewing

Fix and improve this skill with Tessl

tessl review fix ./optional-skills/mlops/obliteratus/SKILL.md

Quality

Content

77%Weight 40%Scale 1-3

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The body is highly actionable with a well-sequenced, validated workflow, but it is long and its progressive disclosure is uneven: a referenced templates directory is missing, one reference file is orphaned, and navigation relies on skill_view calls rather than clear links. Tightening verbose sections and fixing the reference structure would lift the weaker dimensions.

Suggestions

Create the missing templates/ directory referenced by the three YAML config paths, or remove those references.

Either reference references/methods-guide.md in the body or remove the orphaned file; use standard markdown links (e.g. [methods guide](references/methods-guide.md)) instead of skill_view(...) for clearer signaling.

Trim low-value sections (the stats 'What's inside' line, the Video Guide, the longest Common Pitfalls items) or move detailed method/pitfall material into reference files to reduce inline length.

Dimension	Reasoning	Score
Conciseness	The ~340-line body is mostly dense and actionable (commands, parameter tables, code) rather than explaining concepts Claude already knows, but sections like the stats 'What's inside' line, the Video Guide, and a 12-item Common Pitfalls list add length that could be tightened, fitting 'mostly efficient but includes some unnecessary explanation or could be tightened'. Not 1 (no padded conceptual primer); not 3 (not every token earns its place).	2 / 3
Actionability	Copy-paste-ready bash/python blocks (e.g. `obliteratus obliterate <model_name> --method advanced --quantization 4bit`), a flags table with concrete defaults, and runnable verification code give fully executable, specific guidance, matching the 'fully executable code/commands; copy-paste ready' anchor.	3 / 3
Workflow Clarity	A clear 7-step sequence (Install → Hardware → Browse → Choose method → Run → Verify → Use) is paired with explicit validation in Step 6 (refusal rate / perplexity / KL thresholds) and error-recovery feedback loops ('If refusals persist...', 'If coherence is damaged...'), matching 'clear sequence with explicit validation steps; feedback loops for error recovery'. The destructive/batch cap does not apply since verification is present.	3 / 3
Progressive Disclosure	The body pushes the 28 analysis modules to a real references file, but it remains largely monolithic (~340 lines) with inline content that could be split (full CLI reference, 9 methods, 12 pitfalls); navigation uses `skill_view(...)` calls rather than clear markdown links, `references/methods-guide.md` is orphaned, and the three `templates/*.yaml` paths are referenced but do not exist. This fits 'some structure but could be better organized; references present but not clearly signaled; content that should be separate is inline'. Not 3 due to broken/orphaned refs and unclear signaling; not 1 because there is genuine structure and no deep nesting.	2 / 3
	Total	10 / 12 Passed

Description

57%Weight 40%Scale 1-3

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

The description is specific and distinctive but terse: it conveys the core action and niche yet omits an explicit 'when to use' trigger clause and common natural terms like 'uncensor'. Adding a 'Use when...' sentence with broader trigger keywords would raise completeness and trigger_term_quality.

Suggestions

Add an explicit trigger clause, e.g. 'Use when the user wants to uncensor or abliterate an LLM, remove refusal guardrails, or create an uncensored version of Llama, Qwen, or Mistral.'

Include common natural terms users say ('uncensor', 'refusal removal', 'guardrails') alongside the technical 'diff-in-means'.

List one or two more concrete actions (e.g. 'identify and excise refusal directions from model weights') to move specificity from domain-plus-one to multiple specific actions.

Dimension	Reasoning	Score
Specificity	The phrase "abliterate LLM refusals (diff-in-means)" names the domain (LLM refusals) and one concrete action plus technique, but does not list multiple specific actions, matching the 'names domain and some actions, but not comprehensive' anchor. It is not vague (not 1) and not a multi-action list (not 3).	2 / 3
Completeness	It states what the skill does ("abliterate LLM refusals") but has no 'Use when...' clause or equivalent explicit trigger guidance, so per the judging guidelines completeness is capped at 2 ('has what, but when is missing or only implied'). Not 3 (no explicit trigger); not 1 (the 'what' is present).	2 / 3
Trigger Term Quality	"abliterate" and "LLM refusals" are natural user terms, but common variations users actually say (e.g. "uncensor", "refusal removal", "guardrails") are absent and "diff-in-means" is technical jargon, fitting 'some relevant keywords but missing common variations'. Not 3 due to gaps; not 1 since real natural terms are present.	2 / 3
Distinctiveness Conflict Risk	Removing refusal behaviors from LLMs is a clear, narrow niche with distinct triggers unlikely to fire for unrelated skills, matching 'clear niche with distinct triggers; unlikely to conflict'.	3 / 3
	Total	9 / 12 Passed

Validation

81%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 13 / 16 Passed

Validation for skill structure

Criteria	Description	Result
metadata_version	'metadata.version' is missing	Warning
metadata_field	'metadata' should map string keys to string values	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	13 / 16 Passed

Repository: NousResearch/hermes-agent
Path: optional-skills/mlops/obliteratus/SKILL.md
Commit: 4c9628e

Reviewed: 6 days ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.