unsloth

Unsloth: 2-5x faster LoRA/QLoRA fine-tuning, less VRAM.

Quality

46%

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 ./optional-skills/mlops/training/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 SKILL.md body is an unfilled template: its Quick Reference is empty, it contains no code or commands, and it directs readers to reference files that do not exist while omitting the ones that do. Real value is isolated in the references/ bundle, so the body fails as an actionable overview.

Suggestions

Populate the Quick Reference with actual common patterns (e.g. a minimal FastLanguageModel.from_pretrained + get_lora_model_training_args snippet) instead of the placeholder line.

Fix the Reference Files section to list the real files (index.md, llms-txt.md, llms.md, llms-full.md) and remove pointers to nonexistent categories like getting_started, tutorials, api, and guides.

Delete the empty scripts/ and assets/ placeholder sections and the generic Notes/Updating boilerplate to remove pure token padding.

Dimension	Reasoning	Score
Conciseness	The body is dominated by empty boilerplate padding ("Quick reference patterns will be added as you use the skill.", "Add helper scripts here for common automation tasks.", "Add templates, boilerplate, or example projects here.") that conveys nothing, matching noticeably verbose with several padded sections; it avoids a 1 only because it is short and does not explain basic concepts.	2 / 5
Actionability	There is no code, command, or example anywhere in the body; the only guidance is the high-level hint "Use `view` to read specific reference files when detailed information is needed," matching minimal concrete guidance with missing executable steps rather than entirely vague (a real pointer to references/llms-txt.md exists).	2 / 5
Workflow Clarity	"Working with This Skill" offers rough hints (beginners start with getting_started/tutorials; features use api/guides category files) but these files do not exist in references/, and there is no sequenced workflow or validation checkpoint — rough sequence-like hints with many gaps.	2 / 5
Progressive Disclosure	The bundle does contain one-level-deep real reference files (references/index.md, llms-txt.md, llms.md, llms-full.md), but the body lists only llms-txt.md and points readers at nonexistent categories (getting_started, tutorials, api, guides), so structure exists but references are inaccurately signaled and the discovery path is broken.	3 / 5
	Total	9 / 20 Passed

Description

65%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 is concise and distinctive, with concrete quantified claims and natural trigger terms centered on the named framework. Its main weakness is the complete absence of a "Use when..." trigger clause, which caps completeness, and it lists outcomes rather than actionable capabilities.

Suggestions

Add a "Use when..." clause, e.g. "Use when fine-tuning LLMs with Unsloth, or when the user mentions LoRA/QLoRA, VRAM limits, or slow training runs."

Enumerate concrete actions/capabilities (e.g. setting up FastLanguageModel, configuring LoRA adapters, running training, saving to GGUF/merged formats) rather than only outcomes.

Include a few more natural synonyms such as "fine-tuning Llama/Mistral/Gemma/Qwen" to broaden trigger coverage.

Dimension	Reasoning	Score
Specificity	"2-5x faster LoRA/QLoRA fine-tuning, less VRAM" names the domain and gives two concrete claims (speedup, VRAM savings), but lists no actions and is not comprehensive — matching the anchor for domain plus 1-2 concrete items rather than the several specific actions of a 4.	3 / 5
Completeness	It clearly states what the skill delivers (faster LoRA/QLoRA fine-tuning, less VRAM) but contains no "Use when..." or equivalent trigger clause, which per the judging guidelines caps completeness at 3; it is above a 2 because the "what" is concrete, not vague.	3 / 5
Trigger Term Quality	"Unsloth", "LoRA/QLoRA", "fine-tuning", and "VRAM" are terms users would naturally say, but common variations like "training", "PEFT", "adapters", or model-family names are missing, matching good-but-incomplete keyword coverage.	4 / 5
Distinctiveness Conflict Risk	Naming the specific framework "Unsloth" gives it a clear niche with distinct triggers and minimal conflict risk; at most it marginally overlaps other LoRA fine-tuning skills, but the explicit library name keeps it distinguishable.	5 / 5
	Total	15 / 20 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/training/unsloth/SKILL.md
Commit: 68fa7e9

Reviewed: 3 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.