fine-tuning-expert

Use when fine-tuning LLMs, training custom models, or adapting foundation models for specific tasks. Invoke for configuring LoRA/QLoRA adapters, preparing JSONL training datasets, setting hyperparameters for fine-tuning runs, adapter training, transfer learning, finetuning with Hugging Face PEFT, OpenAI fine-tuning, instruction tuning, RLHF, DPO, or quantizing and deploying fine-tuned models. Trigger terms include: LoRA, QLoRA, PEFT, finetuning, fine-tuning, adapter tuning, LLM training, model training, custom model.

Quality

88%

Does it follow best practices?

Impact

—

No eval scenarios have been run

Securityby

Advisory

Suggest reviewing before use

Quality

Content

77%

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

This is a strong, actionable skill with a clear multi-step workflow, explicit validation checkpoints, and fully executable code examples. Its main weaknesses are the missing bundle reference files that the reference table points to, and the somewhat lengthy inline code that could be offloaded to keep the overview leaner. The persona description at the top adds no value for Claude.

Suggestions

Remove the persona description ('Senior ML engineer specializing in...') — it wastes tokens and doesn't change Claude's behavior.

Provide the five referenced files (references/lora-peft.md, etc.) or remove the reference table to avoid dead links that undermine progressive disclosure.

Dimension	Reasoning	Score
Conciseness	The skill is mostly efficient with well-commented code and useful inline guidance, but includes some unnecessary framing (e.g., 'Senior ML engineer specializing in...' persona description) and the constraints section partially restates what's already implied by the workflow. The code example is long but justified given the complexity of the task.	2 / 3
Actionability	Provides a fully executable, copy-paste-ready LoRA fine-tuning example with real library imports, concrete hyperparameter values with explanatory comments, QLoRA variant, and adapter merging code. The output templates section specifies exactly what deliverables to produce.	3 / 3
Workflow Clarity	The 5-step core workflow has explicit validation checkpoints at each stage (dataset validation before training, loss curve monitoring during training, held-out evaluation before deployment). Feedback loops are present — e.g., 'fix all errors before proceeding' and 'plateau or increase signals overfitting' — providing clear error recovery guidance.	3 / 3
Progressive Disclosure	The reference table with 'Load When' conditions is a well-designed progressive disclosure pattern, but no bundle files are provided, meaning all five referenced files (references/lora-peft.md, etc.) are missing. The inline code example is quite long and could arguably be split into a reference file, keeping the SKILL.md as a leaner overview.	2 / 3
	Total	10 / 12 Passed

Description

100%

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

This is a strong skill description that clearly defines its domain (LLM fine-tuning), lists numerous specific actions and techniques, and provides explicit trigger guidance with both a 'Use when' clause and enumerated trigger terms. The description covers natural user language variations alongside technical terminology, making it highly discoverable and unlikely to conflict with other skills.

Dimension	Reasoning	Score
Specificity	Lists multiple specific concrete actions: configuring LoRA/QLoRA adapters, preparing JSONL training datasets, setting hyperparameters, adapter training, transfer learning, quantizing and deploying fine-tuned models, instruction tuning, RLHF, DPO.	3 / 3
Completeness	Clearly answers both 'what' (configuring adapters, preparing datasets, setting hyperparameters, quantizing/deploying models) and 'when' with an explicit 'Use when' clause at the start and an 'Invoke for' clause detailing specific scenarios. Trigger terms are also explicitly listed.	3 / 3
Trigger Term Quality	Excellent coverage of natural terms users would say, including both technical terms (LoRA, QLoRA, PEFT, DPO, RLHF) and natural language variations (finetuning, fine-tuning, adapter tuning, LLM training, model training, custom model). Also mentions specific frameworks like Hugging Face PEFT and OpenAI fine-tuning.	3 / 3
Distinctiveness Conflict Risk	Occupies a clear niche around LLM fine-tuning and adapter training with highly specific trigger terms (LoRA, QLoRA, PEFT, DPO, RLHF) that are unlikely to conflict with general coding, data processing, or other ML skills. The domain is well-scoped.	3 / 3
	Total	12 / 12 Passed

Validation

100%

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 — 11 / 11 Passed

Validation for skill structure

No warnings or errors.

Repository: Jeffallan/claude-skills
Commit: e8be415

Reviewed: 21 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.