CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

ray-data

Scalable data processing for ML workloads. Streaming execution across CPU/GPU, supports Parquet/CSV/JSON/images. Integrates with Ray Train, PyTorch, TensorFlow. Scales from single machine to 100s of nodes. Use for batch inference, data preprocessing, multi-modal data loading, or distributed ETL pipelines.

Quality

75%

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/data-engineering/ray-data/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

57%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 catalog with mostly executable code examples, but it underuses its own bundle: integration and transformation content is fully duplicated between SKILL.md and the references files, and marketing-flavored sections (benchmarks, customer logos, version pins) spend tokens without adding actionable guidance. Batch-style operations also lack any validation checkpoints.

Suggestions

De-duplicate the Ray Train, PyTorch, TensorFlow, and map/filter/groupby/GPU sections against references/integration.md and references/transformations.md — keep one short inline example each in SKILL.md and move the full guides to the reference files.

Remove or relocate non-actionable marketing content (the performance benchmark numbers, Pinterest/ByteDance/Spotify use-case logos, and the Ray 2.40.0+ version pin) to reclaim context-window tokens.

Add a validation step to the batch patterns, e.g. after ds.write_parquet, read back a sample with ray.data.read_parquet(output) and verify row counts/prediction sanity before declaring the run complete.

DimensionReasoningScore

Conciseness

The code examples are lean, but the body pads tokens that earn nothing: a "Performance benchmarks" section with marketing numbers ("1 node (16 cores): ~30 minutes"), a "Use cases" section naming Pinterest/ByteDance/Spotify, and a version pin ("Ray 2.40.0+") that is time-sensitive and not placed in a deprecated/old-patterns section. This fits anchor 3 — mostly efficient with some unnecessary material that should be trimmed.

3 / 5

Actionability

Most examples are executable copy-paste (read_parquet, map_batches with num_gpus, write_parquet, to_torch/to_tf), covering the common cases. Minor gaps keep it below 5: the batch-inference example calls an undefined MyModel() and the pipeline example references undefined clean_data/tokenize/augment functions, so those snippets are not runnable as-is.

4 / 5

Workflow Clarity

Content is organized as a topical catalog rather than a sequenced workflow, and the batch operations it teaches (distributed batch inference, writing output to s3) include no validation or verification steps — no post-write read-back, no sanity check on predictions. Per the rubric, missing validation in batch operations caps workflow_clarity at 3.

3 / 5

Progressive Disclosure

The two reference files are real, one level deep, and clearly signaled in a References section — but the body inlines substantial material that duplicates them: Ray Train/PyTorch/TensorFlow examples appear both in SKILL.md and in references/integration.md, and map/filter/groupby/GPU examples appear both inline and in references/transformations.md. Content that should live in the separate files is inline, which is the anchor-3 pattern.

3 / 5

Total

13

/

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, third-person description that clearly states what the skill does and when to use it, with concrete formats, frameworks, and use-case triggers. It leans slightly on capability adjectives ("Scalable") rather than enumerating explicit operations, and its broad data-processing terms leave small overlap risk with sibling ETL/distributed-data skills.

DimensionReasoningScore

Specificity

Names several concrete capabilities — "Streaming execution across CPU/GPU, supports Parquet/CSV/JSON/images", "Integrates with Ray Train, PyTorch, TensorFlow", "Scales from single machine to 100s of nodes" — plus concrete use-actions ("batch inference, data preprocessing, multi-modal data loading, or distributed ETL pipelines"). Falls short of a 5 because it leads with capability framing ("Scalable data processing for ML workloads") rather than explicit read/write/transform verbs, leaving minor coverage gaps.

4 / 5

Completeness

Explicitly answers both questions: the "what" is stated concretely (streaming execution, supported formats, framework integrations, scaling range) and the "when" is an explicit trigger clause — "Use for batch inference, data preprocessing, multi-modal data loading, or distributed ETL pipelines". This matches the anchor-5 pattern of clear what plus concrete trigger phrases, not the anchor-4 case where 'when' could be more specific.

5 / 5

Trigger Term Quality

Good coverage of natural phrases users would say: "batch inference", "data preprocessing", "multi-modal data loading", "distributed ETL pipelines", plus format and framework names (Parquet, CSV, JSON, PyTorch, TensorFlow). A few natural synonyms/extensions are missing (e.g., "ETL pipeline", "shuffle data", file extensions), so it fits anchor 4 rather than 5.

4 / 5

Distinctiveness Conflict Risk

The ML-workload framing, named frameworks, and Ray Train integration carve out a distinct niche, but "data preprocessing" and "distributed ETL pipelines" remain broadly applicable to sibling data-processing skills (Dask, Spark), giving minor overlap risk — anchor 4 rather than the minimal-conflict anchor 5.

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/data-engineering/ray-data/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.