unsloth

Expert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization

Quality

45%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./skills/vendor-ai-research/unsloth/SKILL.md

Quality

Content

28%Weight 40%Scale 1-5

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

The body is an auto-generated template shell that adds almost no substantive guidance: no code, no workflow, and references that mismatch the actual bundle. Its only virtues are brevity and a real pointer to one reference file.

Suggestions

Add a concrete quick-start code block (e.g. a minimal Unsloth LoRA fine-tuning snippet) so the body provides executable guidance instead of only describing.

Replace the broken/placeholder references (getting_started, tutorials, api, guides) with the real bundle files that exist (index.md, llms-txt.md, llms-full.md, llms.md) and signal them one level deep.

Provide a real multi-step fine-tuning workflow with a validation checkpoint (e.g. load model -> apply LoRA -> train -> evaluate loss) instead of the empty 'Quick reference' placeholder.

Dimension	Reasoning	Score
Conciseness	The body is short and does not over-explain concepts Claude already knows, but contains filler boilerplate ('Quick reference patterns will be added as you use the skill') and generic directory-purpose descriptions that could be trimmed. Not a 4 because of this noticeable filler; not a 2 because it is not heavily verbose or padded with concept explanations.	3 / 5
Actionability	There is no code, no commands, and no concrete examples — the body only describes ('Use the appropriate category reference file') and points abstractly to reference files. Matches anchor 1 (entirely vague/abstract, only describes rather than instructs); not a 2 because even the hints are generic rather than specific steps.	1 / 5
Workflow Clarity	No workflow exists for the core fine-tuning task; the only sequence is a 2-step 'Updating' section, with no validation checkpoints for a batch training operation. Not a 1 because at least one rough sequence is present; not a 3 because the core task has no sequenced steps and validation is entirely absent.	2 / 5
Progressive Disclosure	Section structure is present and one real one-level reference is signaled (llms-txt.md), but the body points to nonexistent files (getting_started, tutorials, api, guides) and omits three of the four actual bundle files (index.md, llms-full.md, llms.md). Not a 4 because references are incomplete and misleading; not a 2 because genuine section structure and a real one-level reference exist.	3 / 5
	Total	9 / 20 Passed

Description

61%Weight 40%Scale 1-5

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 clearly conveys what the skill does and is anchored to a distinct, named library, with solid natural trigger terms. Its main weakness is the absence of an explicit 'Use when...' trigger clause, which caps completeness and leaves the activation guidance implicit.

Suggestions

Append an explicit 'Use when...' clause, e.g. 'Use when fine-tuning Llama/Mistral/Gemma/Qwen models with Unsloth, or when the user wants faster or lower-memory LoRA/QLoRA training.'

Add concrete model-name synonyms (Llama, Mistral, Gemma, Qwen) to broaden natural trigger coverage.

Convert benefit metrics into a couple of concrete action verbs (e.g. 'fine-tune', 'apply LoRA/QLoRA', 'reduce memory') to raise specificity.

Dimension	Reasoning	Score
Specificity	Names the domain and concrete capabilities ('fast fine-tuning', '50-80% less memory', 'LoRA/QLoRA optimization'), but lists benefit metrics rather than a comprehensive set of discrete actions. Not a 4 because it does not enumerate several specific actions; not a 2 because it goes beyond minimal/generic capability statements.	3 / 5
Completeness	Has a clear 'what' (expert guidance for fast fine-tuning with Unsloth) but no 'Use when...' clause or equivalent explicit trigger guidance, capping completeness at 3 per the rubric guidelines. Not a 4 because 'when' is entirely missing rather than weakly present.	3 / 5
Trigger Term Quality	Includes natural terms users would say ('fine-tuning', 'Unsloth', 'LoRA', 'QLoRA', 'training', 'memory') with good coverage. Not a 5 because common model-name synonyms (Llama, Mistral, Gemma, Qwen) and variations are absent; not a 3 because keyword coverage is genuinely strong rather than partial.	4 / 5
Distinctiveness Conflict Risk	Unsloth is a specific named library giving it a clear niche, with minor overlap risk only against general PEFT/fine-tuning skills. Not a 5 because LoRA/QLoRA optimization could still overlap with generic fine-tuning skills; not a 3 because the named tool makes it mostly distinct.	4 / 5
	Total	14 / 20 Passed

Validation

93%

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 — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: OpenRaiser/NanoResearch
Path: skills/vendor-ai-research/unsloth/SKILL.md
Commit: 9d3b440

Reviewed: 10 hours 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.