CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

databricks-ml-training

Train ML models on Databricks. Use for: classification/regression/deep-learning (XGBoost, scikit-learn, LightGBM, PyTorch) with Optuna, @prod/@challenger aliases, batch scoring (spark_udf for plain models, fe.score_batch for feature-store-backed), custom PyFunc, custom ResponsesAgent (LangGraph + UC Function/Vector Search); UC feature tables + FeatureLookup + point-in-time joins + Lakebase online store; declarative Feature Views (create_feature, DeltaTableSource, RollingWindow/SlidingWindow/TumblingWindow, materialize_features, streaming Kafka features). NOT for: endpoint ops (databricks-model-serving), MLflow evaluation (databricks-mlflow-evaluation).

Quality

92%

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

SKILL.md
Quality
Evals
Security

Quality

Content

92%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.

An exemplary operational skill body: executable end-to-end code, a validated async-job workflow with error recovery, and clean progressive disclosure to real, well-signaled reference files. The only trimmable fat is the repeated fe.score_batch and model-serving cross-references, which costs it a perfect conciseness score.

DimensionReasoningScore

Conciseness

The body is dense and operational — it assumes competence ('For imbalanced labels: stratify the split, set scale_pos_weight = neg/pos') and never explains what MLflow or XGBoost is — but repeats the fe.score_batch-vs-spark_udf divergence in at least four places (consumption table, callout after the flow diagram, Consume section intro, Feature Engineering section) and cross-references databricks-model-serving five times. Fits the level-4 anchor (efficient, minor instances that could be trimmed) rather than level 5, where every token would earn its place.

4 / 5

Actionability

Fully executable guidance throughout: a complete notebook (imports, registry URI setup, Optuna objective, retrain-and-register, alias repoint), copy-paste-ready batch scoring code with env_manager rules, and a complete bash job-submit sequence including the jq polling loop, result_state check, and TASK run_id output fetch. Traps are stated as concrete fixes ('environments[].spec.client: "4" is required', 'use the TASK run_id — NOT the submit run_id'). Matches level 5; the only placeholder (X_train/y_train) is dataset-specific by nature and sits after an explicit data-exploration cell.

5 / 5

Workflow Clarity

The multi-step flow is clearly sequenced (canonical ASCII diagram silver → train/register/alias → score → gold, then sections in execution order) and the risky batch operation (serverless job) has explicit validation with feedback loops: poll to terminal life_cycle_state, cancel-run + exit 1 on timeout, check result_state == SUCCESS before proceeding, then pull structured output via the TASK run_id. Matches the level-5 anchor (explicit validation steps and error-recovery loops for a batch operation).

5 / 5

Progressive Disclosure

The body is a genuine overview: the 90% case is inline while deep topics are split into four real, one-level-deep reference files (custom-pyfunc.md, genai-agents.md, feature-store.md, feature-views.md — all verified present, including the genai-agents.md#resources-that-need-passthrough-auth anchor). Each reference is signaled inline at its point of use with a 'when to reach for it' heuristic and summarized again in a Reference files table, plus a Related skills section. Matches the level-5 anchor for clear overview with well-signaled, appropriately split references.

5 / 5

Total

19

/

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 highly specific, well-scoped description that answers what, when, and when-not with concrete capability lists and named technologies. Its only weakness is that a large share of trigger vocabulary is API-level jargon rather than natural user phrasing, leaving a few common synonyms uncovered.

DimensionReasoningScore

Specificity

Lists multiple concrete actions with comprehensive coverage: 'classification/regression/deep-learning (XGBoost, scikit-learn, LightGBM, PyTorch) with Optuna', 'batch scoring (spark_udf for plain models, fe.score_batch for feature-store-backed)', 'custom PyFunc', 'custom ResponsesAgent (LangGraph + UC Function/Vector Search)', plus Feature Views specifics ('create_feature, DeltaTableSource, RollingWindow/SlidingWindow/TumblingWindow, materialize_features'). Matches the level-5 anchor of multiple specific concrete actions; not level 4 because coverage spans training, scoring, serving, and feature engineering with named APIs rather than having minor gaps.

5 / 5

Completeness

Explicitly answers both: 'what' via 'Train ML models on Databricks' and 'when' via the explicit 'Use for: classification/regression/deep-learning...' trigger clause with concrete capabilities. Exceeds level 4 because the 'when' is explicit and itemized, and the 'NOT for:' clause adds negative trigger guidance; not below 5 since neither element is missing or vague.

5 / 5

Trigger Term Quality

Good natural-term coverage ('Train ML models', 'classification', 'regression', 'XGBoost', 'PyTorch', 'batch scoring', 'feature store', 'feature tables') but heavy on API-level jargon ('@prod/@challenger aliases', 'DeltaTableSource', 'RollingWindow/SlidingWindow/TumblingWindow') that users rarely say verbatim, and natural synonyms like 'hyperparameter tuning', 'retraining', or 'model registry' are absent. Fits the level-4 anchor (good coverage, a few natural terms missing) rather than level 5, which requires comprehensive synonyms.

4 / 5

Distinctiveness Conflict Risk

Clear niche (ML training on Databricks) with explicit de-scoping of adjacent skills: 'NOT for: endpoint ops (databricks-model-serving), MLflow evaluation (databricks-mlflow-evaluation)'. Distinct trigger terms (training frameworks, feature engineering) minimize conflict risk; this is stronger than the level-4 'minor overlap risk' anchor because boundaries are explicitly drawn.

5 / 5

Total

19

/

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

frontmatter_unknown_keys

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

Warning

relative_links

Relative link issues: 12 suspicious

Warning

Total

14

/

16

Passed

Repository

databricks/databricks-agent-skills

Path

plugins/databricks/claude/skills/databricks-ml-training/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.