Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →No New U-Net — self-configuring framework for medical image segmentation. Automatically adapts to any dataset. Top performer on biomedical segmentation benchmarks (BraTS, KiTS, etc.).
.claude/skills/mkurman-nnunet/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | -5% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | -60% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | -42% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | -32% | 0% |
nnUNet (No New U-Net) is a self-configuring framework for medical image segmentation that automatically adapts to any dataset. Consistently top-performing on benchmarks like BraTS, KiTS, and AMOS.
bashuv pip install nnunetv2
bashnnUNetv2_plan_and_preprocess -d DATASET_ID -pl nnUNetPlanner
bashnnUNetv2_train DATASET_ID CONFIG 0 # CONFIG: 2d, 3d_fullres, 3d_lowres
bashnnUNetv2_predict -i INPUT_FOLDER -o OUTPUT_FOLDER -d DATASET_ID -c CONFIG
pythonfrom nnunetv2.inference.predict_from_raw_data import nnUNetPredictor predictor = nnUNetPredictor() predictor.initialize_from_trained_model_folder("nnUNet_results/DatasetXYZ", "3d_fullres") predictor.predict_from_files("input_images", "output_segmentations")
nnUNetv2_plan_and_preprocess for automatic configurationnnUNetv2_trainnnUNetv2_predict or Python API| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 5,787 | 2,893 | -50% | 1 | 1 | 0% | 1,128 | 856 | -24% | 0 | 0 | — |
case-02 | pass→pass | 7,614 | 5,396 | -29% | 1 | 1 | 0% | 1,414 | 1,338 | -5% | 0 | 0 | — |
case-03 | fail→fail | 11,398 | 6,634 | -42% | 1 | 1 | 0% | 2,103 | 1,559 | -26% | 0 | 0 | — |
case-04 | pass→pass | 9,735 | 2,377 | -76% | 1 | 1 | 0% | 1,762 | 710 | -60% | 0 | 0 | — |
case-05 | pass→pass | 11,837 | 4,990 | -58% | 1 | 1 | 0% | 2,281 | 1,316 | -42% | 0 | 0 | — |
case-06 | pass→pass | 15,536 | 3,859 | -75% | 1 | 1 | 0% | 1,587 | 1,080 | -32% | 0 | 0 | — |
case-07 | pass→pass | 9,440 | 5,128 | -46% | 1 | 1 | 0% | 1,741 | 1,335 | -23% | 0 | 0 | — |
case-08 | pass→pass | 11,706 | 7,569 | -35% | 1 | 1 | 0% | 1,937 | 1,685 | -13% | 0 | 0 | — |
case-09 | pass→pass | 7,503 | 2,665 | -64% | 1 | 1 | 0% | 1,379 | 835 | -39% | 0 | 0 | — |
case-10 | pass→pass | 4,110 | 2,435 | -41% | 1 | 1 | 0% | 719 | 726 | +1% | 0 | 0 | — |
case-11 | pass→pass | 13,357 | 10,647 | -20% | 1 | 1 | 0% | 2,174 | 1,758 | -19% | 0 | 0 | — |
case-12 | pass→pass | 8,062 | 5,041 | -37% | 1 | 1 | 0% | 1,608 | 1,297 | -19% | 0 | 0 | — |
case-13 | pass→pass | 4,540 | 3,000 | -34% | 1 | 1 | 0% | 859 | 892 | +4% | 0 | 0 | — |
case-14 | pass→pass | 3,834 | 1,814 | -53% | 1 | 1 | 0% | 595 | 611 | +3% | 0 | 0 | — |
case-15 | fail→fail | 10,503 | 8,628 | -18% | 1 | 1 | 0% | 1,983 | 1,995 | +1% | 0 | 0 | — |
case-16 | pass→pass | 7,594 | 5,066 | -33% | 1 | 1 | 0% | 1,392 | 1,264 | -9% | 0 | 0 | — |
case-17 | pass→pass | 17,784 | 17,593 | -1% | 1 | 1 | 0% | 3,452 | 3,694 | +7% | 0 | 0 | — |
case-18 | pass→pass | 16,353 | 17,495 | +7% | 1 | 1 | 0% | 3,105 | 3,996 | +29% | 0 | 0 | — |
case-19 | pass→pass | 13,415 | 8,532 | -36% | 1 | 1 | 0% | 2,480 | 1,993 | -20% | 0 | 0 | — |
case-20 | pass→pass | 4,761 | 3,559 | -25% | 1 | 1 | 0% | 858 | 953 | +11% | 0 | 0 | — |
case-21 | pass→pass | 12,086 | 7,589 | -37% | 1 | 1 | 0% | 2,186 | 1,784 | -18% | 0 | 0 | — |
case-22 | pass→pass | 5,342 | 2,586 | -52% | 1 | 1 | 0% | 931 | 767 | -18% | 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.