ray-train

Distributed training orchestration across clusters. Scales PyTorch/TensorFlow/HuggingFace from laptop to 1000s of nodes. Built-in hyperparameter tuning with Ray Tune, fault tolerance, elastic scaling. Use when training massive models across multiple machines or running distributed hyperparameter sweeps.

Quality

82%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Quality

Content

65%

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

Highly actionable body with many executable examples, but it is let down by time-sensitive claims outside a deprecated section, missing validation checkpoints for a batch operation, and two broken reference-file links. Structurally sound yet not fully lean or navigable.

Suggestions

Create the missing referenced files references/hyperparameter-tuning.md and references/custom-loops.md, or remove their links from the 'Advanced topics' section, so all referenced paths resolve.

Move time-sensitive values ('Version: 2.40.0+', '⭐ 36,000+') into a clearly labeled version/maintenance section and drop the redundant restatement of Ray Train's purpose, since the description already covers it.

Add an explicit validation/verification checkpoint to the training workflows (e.g., check ray status and confirm workers are Ready before trainer.fit(), and verify checkpoint integrity on resume) to satisfy the batch-operation feedback-loop requirement.

Dimension	Reasoning	Score
Conciseness	Mostly efficient code-driven content, but time-sensitive claims appear outside a deprecated section ('Version: 2.40.0+', 'GitHub ⭐ 36,000+'), the opener restates what Ray Train does, and Quick start overlaps with Workflow 1. Not score 3 because these tokens do not all earn their place; not score 1 because the bulk is concrete rather than padded conceptual explanation.	2 / 3
Actionability	Provides multiple fully executable, copy-paste-ready examples (PyTorch, HuggingFace, Ray Tune, checkpointing, multi-node) plus concrete shell commands ('ray start --head --port=6379', 'pip install -U "ray[train]"'). Not score 2 because the guidance is concrete and complete rather than pseudocode or abstract.	3 / 3
Workflow Clarity	Workflows are sequenced as recipes, but distributed training is a long-running batch operation and no explicit validate→fix→retry checkpoint is given, which caps this dimension at 2 per the rubric. Not score 3 because feedback loops are absent; not score 1 because steps are clearly listed rather than missing.	2 / 3
Progressive Disclosure	The 'Advanced topics' section defers to one-level-deep, well-signaled links, but only 1 of 3 referenced files exists (multi-node.md present; hyperparameter-tuning.md and custom-loops.md missing), and all five workflows are inline rather than split out. Not score 3 because navigation is broken by the missing files and content is not appropriately split; not score 1 because structure and signaling are present and references are only one level deep.	2 / 3
	Total	9 / 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.

A strong, third-person description that names concrete capabilities, includes explicit 'Use when' trigger guidance, and occupies a distinct multi-node distributed-training niche. All four dimensions land at the top of the scale.

Dimension	Reasoning	Score
Specificity	Names multiple concrete capabilities — 'Distributed training orchestration across clusters', 'Scales PyTorch/TensorFlow/HuggingFace from laptop to 1000s of nodes', 'hyperparameter tuning with Ray Tune, fault tolerance, elastic scaling' — rather than vague language. Voice is third-person ('Scales', 'Built-in'), so no first/second-person penalty applies; not score 2 because it goes well beyond naming a domain.	3 / 3
Completeness	Explicitly answers what ('Distributed training orchestration... Built-in hyperparameter tuning... fault tolerance, elastic scaling') and when ('Use when training massive models across multiple machines or running distributed hyperparameter sweeps'). Not score 2 because the 'Use when' trigger clause is present and explicit.	3 / 3
Trigger Term Quality	Includes natural terms a user would say — 'distributed training', 'PyTorch/TensorFlow/HuggingFace', 'hyperparameter tuning', 'training massive models across multiple machines', 'distributed hyperparameter sweeps'. Not score 2 because coverage spans the common variations users actually request.	3 / 3
Distinctiveness Conflict Risk	Occupies a clear niche (multi-node distributed training orchestration with Ray) with triggers tied to multi-machine training and distributed sweeps that are unlikely to fire for single-machine skills. Not score 2 because the framing is specific rather than generically overlapping.	3 / 3
	Total	12 / 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
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning
relative_links	Relative link issues: 2 missing	Warning
referenced_paths_exist	Referenced path issues: 4 missing	Warning

	Total	13 / 16 Passed

Repository: Orchestra-Research/AI-Research-SKILLs
Path: 08-distributed-training/ray-train/SKILL.md
Commit: 773a529

Reviewed: about 13 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.