Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Spatial analyst
.claude/skills/spatial-transcriptomics-agent/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | -15% | 0% |
| case-01 | ✓→✗ | ▼ Worse | -26% | 0% |
| case-02 | ✓→✗ | ▼ Worse | -29% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 23% | 0% |
| case-07 | ✓→✓ | ＝ Same ✓ | -8% | 0% |
<!--
#
#
-->
Run STAgent to align histology images with expression matrices, perform clustering/SVG detection, and generate literature-backed spatial reports.
conda env create -f environment.yml && conda activate STAgent.expression_path (.h5ad/Spaceranger) + image_path (H&E/IF) and metadata.cluster, find_svg, annotate_domains, or composite instructions; run python repo/src/main.py --data_path ... --task "...".textUser: "Analyze this breast cancer ST dataset, find immune infiltrates." Agent: loads data, runs `sqidpy.gr.spatial_neighbors`, computes Leiden clusters, plots marker genes (CD3D, CD19), and summarizes which clusters map to tumor core vs. stromal/immune zones.
README.md for detailed instructions.<!-- AUTHOR_SIGNATURE: 9a7f3c2e-MD-BABU-MIA-2026-MSSM-SECURE -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→fail | 23,784 | 16,129 | -32% | 1 | 1 | 0% | 5,120 | 3,792 | -26% | 0 | 0 | — |
case-06 | pass→pass | 13,943 | 15,155 | +9% | 1 | 1 | 0% | 2,348 | 2,891 | +23% | 0 | 0 | — |
case-07 | pass→pass | 12,288 | 7,923 | -36% | 1 | 1 | 0% | 2,199 | 2,018 | -8% | 0 | 0 | — |
case-02 | pass→fail | 31,918 | 21,314 | -33% | 1 | 1 | 0% | 6,197 | 4,372 | -29% | 0 | 0 | — |
case-03 | fail→pass | 40,857 | 21,701 | -47% | 1 | 1 | 0% | 6,194 | 5,260 | -15% | 0 | 0 | — |
case-04 | pass→pass | 14,648 | 10,593 | -28% | 1 | 1 | 0% | 2,408 | 2,214 | -8% | 0 | 0 | — |
case-05 | pass→pass | 10,952 | 8,858 | -19% | 1 | 1 | 0% | 2,073 | 2,185 | +5% | 0 | 0 | — |
case-08 | pass→pass | 16,212 | 10,176 | -37% | 1 | 1 | 0% | 2,900 | 2,350 | -19% | 0 | 0 | — |
case-09 | pass→pass | 6,263 | 4,274 | -32% | 1 | 1 | 0% | 1,003 | 1,325 | +32% | 0 | 0 | — |
case-10 | pass→pass | 10,389 | 1,692 | -84% | 1 | 1 | 0% | 1,564 | 815 | -48% | 0 | 0 | — |
case-11 | pass→pass | 14,414 | 7,386 | -49% | 1 | 1 | 0% | 2,467 | 1,982 | -20% | 0 | 0 | — |
case-12 | pass→pass | 14,331 | 8,590 | -40% | 1 | 1 | 0% | 2,230 | 1,986 | -11% | 0 | 0 | — |
case-13 | pass→pass | 14,534 | 9,278 | -36% | 1 | 1 | 0% | 2,239 | 2,222 | -1% | 0 | 0 | — |
case-14 | pass→pass | 13,907 | 8,771 | -37% | 1 | 1 | 0% | 2,241 | 2,080 | -7% | 0 | 0 | — |
case-15 | pass→pass | 9,145 | 1,903 | -79% | 1 | 1 | 0% | 1,540 | 935 | -39% | 0 | 0 | — |
case-16 | fail→fail | 12,129 | 5,844 | -52% | 1 | 1 | 0% | 1,986 | 1,625 | -18% | 0 | 0 | — |
case-17 | pass→pass | 13,321 | 2,790 | -79% | 1 | 1 | 0% | 2,156 | 1,012 | -53% | 0 | 0 | — |
case-18 | pass→pass | 14,546 | 11,919 | -18% | 1 | 1 | 0% | 2,528 | 2,693 | +7% | 0 | 0 | — |
case-19 | pass→pass | 14,703 | 10,877 | -26% | 1 | 1 | 0% | 2,671 | 2,616 | -2% | 0 | 0 | — |
case-20 | pass→pass | 13,818 | 10,850 | -21% | 1 | 1 | 0% | 2,856 | 2,906 | +2% | 0 | 0 | — |
case-21 | pass→pass | 19,480 | 15,980 | -18% | 1 | 1 | 0% | 3,627 | 3,758 | +4% | 0 | 0 | — |
case-22 | pass→pass | 12,770 | 11,535 | -10% | 1 | 1 | 0% | 2,719 | 3,104 | +14% | 0 | 0 | — |
case-23 | pass→pass | 19,676 | 15,133 | -23% | 1 | 1 | 0% | 3,816 | 3,387 | -11% | 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. 23 cases were attempted. The headline lift of -33 percentage points is the difference between those two pass rates over the 23 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 7/26/2026 | +45% |
Other measured skills in the registry, with their headline benchmark lift.