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

79%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./backend/cli/skills/ml-training/ray-train/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

67%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 a well-organized, largely executable guide with a correctly structured bundle reference and useful troubleshooting. Its main weaknesses are marketing-fluff sections and duplication that waste tokens, plus minor gaps in code completeness (missing imports, undefined helper functions) and absent pre-launch validation checkpoints.

Suggestions

Cut the marketing and social-proof content ('Key advantages' claims, 'That's it!' list, GitHub stars, 'Used by' logos) and fold the version pin into a maintenance note — these pad the body without adding actionable guidance.

Deduplicate the Quick start against Workflow 1 (they show nearly the same scaling transformation) and add the missing imports/definitions (ScalingConfig, train, train_epoch/train_func) so each workflow snippet is self-contained.

Move the manual 'ray start --head/--address' cluster commands out of Workflow 5 into the existing references/multi-node.md, and add an explicit validation step before multi-node training (verify 'ray status' shows all nodes/GPUs ready before calling trainer.fit()).

DimensionReasoningScore

Conciseness

The body is dominated by useful code, but includes padded sections Claude does not need: 'That's it! Ray handles:', 'Key advantages' with marketing claims ('Easiest multi-node setup', 'Best-in-class hyperparameter tuning'), and a 'Resources' section with social proof ('GitHub ⭐ 36,000+', 'Used by: OpenAI, Uber, Spotify...'). Workflow 1 largely duplicates the Quick start example, and the version pin '2.40.0+' is time-sensitive without a deprecation context.

3 / 5

Actionability

Mostly executable, concrete code: the Quick start is copy-paste runnable, and cluster launch commands ('ray start --head --port=6379') are specific. Minor gaps: Workflow 3/4 use ScalingConfig, TorchTrainer, and train without importing them, and Workflows 3-5 call undefined helpers (train_epoch, train_func, MyModel, dataloader).

4 / 5

Workflow Clarity

Clear sequencing: install → quick start → common workflows → troubleshooting → advanced pointer, with an error-recovery path in 'Ray cluster not connecting' (check 'ray status' output shows nodes/GPUs, restart head and workers) and checkpoint/resume logic for failures. Not a 5 because there is no explicit checkpoint before launching multi-node training (e.g., verify workers connected before trainer.fit()) and the OOM/slow-training fixes lack verification steps.

4 / 5

Progressive Disclosure

Good structure with one clearly signaled, one-level-deep reference: 'See [references/multi-node.md](references/multi-node.md) for Ray cluster deployment on AWS, GCP, Kubernetes, and SLURM' — the file exists in the bundle. Not a 5 because Workflow 5 inlines the manual cluster launch commands ('ray start --head --port=6379') that duplicate the multi-node reference, and the four code-heavy workflows could themselves be split out of SKILL.md.

4 / 5

Total

15

/

20

Passed

Description

83%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 description that explicitly states both capabilities and use-when triggers in third person, with good natural keyword coverage. It is comprehensive and specific, with only minor gaps in trigger variations and a small risk of overlapping generic model-training skills.

DimensionReasoningScore

Specificity

Phrases like 'Scales PyTorch/TensorFlow/HuggingFace from laptop to 1000s of nodes', 'hyperparameter tuning with Ray Tune, fault tolerance, elastic scaling' list several concrete capabilities in third person. Not a 5 because 'orchestration across clusters' is somewhat abstract and other capabilities from the body (checkpointing, framework-agnostic training) are omitted.

4 / 5

Completeness

The 'what' is explicit and concrete (scaling, tuning, fault tolerance, elastic scaling) and the 'when' is an explicit 'Use when training massive models across multiple machines or running distributed hyperparameter sweeps' clause with concrete trigger phrases. Both are clearly and explicitly answered.

5 / 5

Trigger Term Quality

Natural phrases present: 'distributed training', 'training massive models across multiple machines', 'distributed hyperparameter sweeps', plus framework names (PyTorch, TensorFlow, HuggingFace, Ray). Missing common variations users would say such as 'multi-GPU', 'scale up training', or 'parallel training'.

4 / 5

Distinctiveness Conflict Risk

The multi-machine and distributed-sweep triggers carve out a clear niche distinct from single-node training skills. Minor overlap risk remains because 'training massive models' alone could also match a generic ML-training or single-node scaling skill.

4 / 5

Total

17

/

20

Passed

Validation

87%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 14 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

metadata_version

'metadata.version' is missing

Warning

frontmatter_unknown_keys

Unknown frontmatter key(s) found; consider removing or moving to metadata

Warning

Total

14

/

16

Passed

Repository

synthetic-sciences/openscience

Path

backend/cli/skills/ml-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.