CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

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

80%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

Fix and improve this skill with Tessl

tessl review fix ./skills/distributed-training/ray-train/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

63%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.

A largely actionable, well-sequenced skill body built around five concrete code workflows plus troubleshooting, held back by promotional padding and two dangling reference files. The main fixes are trimming marketing content and either creating or removing the missing reference files.

Suggestions

Create references/hyperparameter-tuning.md and references/custom-loops.md (or remove those Advanced topics links), since two of the three referenced files are absent from the bundle and currently dead-end the reader.

Trim promotional padding — the "Key advantages" marketing list, "Used by: OpenAI, Uber, Spotify...", and the GitHub star count — to raise token efficiency without losing any actionable content.

Fix the Workflow 4 checkpoint example to use a correct Checkpoint API pattern (e.g., saving to a temp directory and calling train.report(..., checkpoint=Checkpoint.from_directory(...))) so the fault-tolerance workflow is copy-paste runnable.

DimensionReasoningScore

Conciseness

The bulk is dense, useful code, but there is a recurring layer of promotional padding: "That's it! Ray handles:", "Key advantages: ... Best-in-class", "Easiest multi-node setup", "Used by: OpenAI, Uber, Spotify, Shopify, Instacart", and a GitHub star count in Resources. This matches "mostly efficient but includes some unnecessary explanation or could be tightened"; it is not a 4 because the marketing-flavored sections (advantages, adopters, star counts) are clearly trimmable rather than minor.

3 / 5

Actionability

Nearly every section is executable, copy-paste-oriented code with concrete commands (pip install, ray start --head, ray up cluster.yaml, tune.Tuner with ASHA). Minor gaps keep it below 5: the checkpointing example uses questionable API (`Checkpoint.from_directory(...)` followed by `checkpoint.path / "model.pt"`), Workflow 2 builds a standard HuggingFace Trainer inside train_func in a way that would not actually distribute, and several snippets call undefined placeholders (MyModel, train_epoch, train_dataset).

4 / 5

Workflow Clarity

The five workflows are clearly labeled and consistently sequenced (define train_func → prepare model/data → report metrics → construct trainer → fit), and the Common issues section provides a check-diagnose-retry loop for cluster connection. It falls short of 5 because validation checkpoints are implicit rather than explicit — e.g., no verification that the cluster reports the expected GPUs before launching a large run, and no guidance on confirming checkpoint integrity after resume.

4 / 5

Progressive Disclosure

Structure is good on paper — an Advanced topics section with one-level-deep pointers, and references/multi-node.md exists and is well-signaled. But scoring against the actual bundle, two of the three referenced files (references/hyperparameter-tuning.md and references/custom-loops.md) do not exist, so navigation breaks for those topics; this lands at "some structure but could be better organized" rather than the good-structure anchor of 4.

3 / 5

Total

14

/

20

Passed

Description

92%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.

A strong, well-structured description: concrete capabilities, an explicit two-trigger "Use when" clause, good natural keyword coverage, and a distinct multi-machine training niche. Only minor improvements possible by adding a few more user-side synonyms (multi-GPU, cluster training).

DimensionReasoningScore

Specificity

The description lists multiple concrete capabilities: "Scales PyTorch/TensorFlow/HuggingFace from laptop to 1000s of nodes. Built-in hyperparameter tuning with Ray Tune, fault tolerance, elastic scaling." This matches the anchor for multiple specific concrete actions with comprehensive coverage (naming the frameworks, the tuning tool, and resilience features). It is not a 4 because coverage of the skill's actual feature set is essentially complete rather than having notable gaps.

5 / 5

Completeness

It explicitly answers "what" (distributed training orchestration, tuning with Ray Tune, fault tolerance, elastic scaling) and "when" ("Use when training massive models across multiple machines or running distributed hyperparameter sweeps") with concrete trigger phrases, matching the anchor for clearly and explicitly answering both. Not a 4 because the "when" clause is specific and multi-trigger, not merely serviceable.

5 / 5

Trigger Term Quality

Natural phrases users would say are present: "distributed training", "training massive models across multiple machines", "distributed hyperparameter sweeps", plus framework names PyTorch/TensorFlow/HuggingFace and "Ray Tune". However, common variations like "multi-node", "multi-GPU", "cluster", or "scaling training" are missing, which keeps it just below the comprehensive-synonyms anchor of 5.

4 / 5

Distinctiveness Conflict Risk

It occupies a clear niche — distributed multi-machine training orchestration — with named frameworks (Ray Tune, PyTorch/TensorFlow/HuggingFace) and multi-node triggers, so risk of firing for an unrelated skill is minimal. Single-node library questions (e.g., plain PyTorch debugging) would not naturally match the multi-machine/sweep triggers.

5 / 5

Total

19

/

20

Passed

Validation

81%

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

CriteriaDescriptionResult

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

OpenLAIR/dr-claw

Path

skills/distributed-training/ray-train/SKILL.md

Reviewed

Table of Contents

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.