Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →MLflow — open-source MLOps platform. Experiment tracking, model registry, packaging, deployment, and evaluation. Multi-cloud ML workflows with reproducible runs and artifact logging.
.claude/skills/mkurman-mlflow/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | 23% | 0% |
| case-07 | ✓→✓ | ＝ Same ✓ | -33% | 0% |
| case-01 | ✓→✓ | ＝ Same ✓ | 88% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | -6% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | -30% | 0% |
MLflow is the leading open-source MLOps platform covering experiment tracking, model registry, packaging (MLflow Models format), and deployment (MLflow Serving). Supports PyTorch, TensorFlow, scikit-learn, ONNX, XGBoost, and custom models across cloud and on-prem.
bashuv pip install mlflow
pythonimport mlflow mlflow.set_experiment("my_project") with mlflow.start_run(run_name="experiment_1"): mlflow.log_param("learning_rate", 0.01) mlflow.log_param("batch_size", 32) mlflow.log_metric("accuracy", 0.92) mlflow.log_metric("loss", 0.35) mlflow.log_artifact("model.pth") mlflow.pytorch.log_model(model, "model")
pythonmlflow.register_model("runs:/<run_id>/model", "MyModel")
bashmlflow models serve --model-uri models:/MyModel/1 --port 5001 mlflow ui --host 0.0.0.0 --port 5000
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | pass→pass | 5,650 | 2,361 | -58% | 1 | 1 | 0% | 1,028 | 687 | -33% | 0 | 0 | — |
case-01 | pass→pass | 2,035 | 1,389 | -32% | 1 | 1 | 0% | 285 | 537 | +88% | 0 | 0 | — |
case-02 | pass→pass | 5,173 | 3,596 | -30% | 1 | 1 | 0% | 1,012 | 951 | -6% | 0 | 0 | — |
case-03 | pass→pass | 8,899 | 5,148 | -42% | 1 | 1 | 0% | 1,751 | 1,227 | -30% | 0 | 0 | — |
case-04 | pass→pass | 7,028 | 3,014 | -57% | 1 | 1 | 0% | 1,388 | 821 | -41% | 0 | 0 | — |
case-05 | pass→pass | 4,859 | 2,603 | -46% | 1 | 1 | 0% | 862 | 743 | -14% | 0 | 0 | — |
case-06 | pass→pass | 3,595 | 1,940 | -46% | 1 | 1 | 0% | 651 | 634 | -3% | 0 | 0 | — |
case-08 | fail→fail | 3,579 | 2,433 | -32% | 1 | 1 | 0% | 549 | 697 | +27% | 0 | 0 | — |
case-09 | pass→pass | 5,558 | 3,559 | -36% | 1 | 1 | 0% | 946 | 903 | -5% | 0 | 0 | — |
case-10 | pass→pass | 3,190 | 2,396 | -25% | 1 | 1 | 0% | 547 | 734 | +34% | 0 | 0 | — |
case-11 | pass→pass | 8,415 | 5,628 | -33% | 1 | 1 | 0% | 1,593 | 1,292 | -19% | 0 | 0 | — |
case-12 | pass→pass | 2,349 | 2,108 | -10% | 1 | 1 | 0% | 352 | 646 | +84% | 0 | 0 | — |
case-13 | pass→pass | 3,927 | 2,051 | -48% | 1 | 1 | 0% | 638 | 655 | +3% | 0 | 0 | — |
case-14 | fail→pass | 3,068 | 2,224 | -28% | 1 | 1 | 0% | 551 | 676 | +23% | 0 | 0 | — |
case-15 | pass→pass | 2,430 | 1,406 | -42% | 1 | 1 | 0% | 386 | 545 | +41% | 0 | 0 | — |
case-16 | pass→pass | 2,241 | 1,766 | -21% | 1 | 1 | 0% | 348 | 577 | +66% | 0 | 0 | — |
case-17 | pass→pass | 6,939 | 5,403 | -22% | 1 | 1 | 0% | 1,318 | 1,261 | -4% | 0 | 0 | — |
case-18 | pass→pass | 14,779 | 12,060 | -18% | 1 | 1 | 0% | 2,569 | 2,470 | -4% | 0 | 0 | — |
case-19 | pass→pass | 2,849 | 1,757 | -38% | 1 | 1 | 0% | 431 | 588 | +36% | 0 | 0 | — |
case-20 | pass→pass | 10,794 | 8,182 | -24% | 1 | 1 | 0% | 2,056 | 1,950 | -5% | 0 | 0 | — |
case-21 | pass→pass | 3,781 | 3,008 | -20% | 1 | 1 | 0% | 710 | 846 | +19% | 0 | 0 | — |
case-22 | pass→pass | 14,933 | 8,971 | -40% | 1 | 1 | 0% | 2,818 | 2,068 | -27% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +5 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.