Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Milvus — cloud-native vector database for billion-scale similarity search. GPU-accelerated indexing, hybrid search, multi-vector, streaming, and time travel. Distributed deployment with Kubernetes.
.claude/skills/mkurman-milvus/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -61% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -61% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -68% | 0% |
Milvus is a cloud-native vector database for billion-scale similarity search. Supports GPU-accelerated indexing (IVF, HNSW, DiskANN), hybrid search (dense + sparse), multi-vector, streaming ingestion, time travel, and distributed deployment with Kubernetes.
bashdocker compose -f https://github.com/milvus-io/milvus/releases/latest/download/milvus-standalone-docker-compose.yml up -d
pythonfrom pymilvus import connections, Collection, FieldSchema, CollectionSchema, DataType connections.connect(host="localhost", port=19530) schema = CollectionSchema([ FieldSchema("id", DataType.INT64, is_primary=True), FieldSchema("embedding", DataType.FLOAT_VECTOR, dim=384), FieldSchema("text", DataType.VARCHAR, max_length=1000), ]) collection = Collection("documents", schema) collection.create_index("embedding", {"index_type": "IVF_FLAT", "metric_type": "L2", "params": {"nlist": 128}}) collection.load()
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 7,947 | 1,979 | -75% | 1 | 1 | 0% | 1,427 | 562 | -61% | 0 | 0 | — |
case-02 | fail→fail | 8,116 | 8,792 | +8% | 1 | 1 | 0% | 1,756 | 2,146 | +22% | 0 | 0 | — |
case-03 | fail→fail | 12,979 | 9,900 | -24% | 1 | 1 | 0% | 2,660 | 2,256 | -15% | 0 | 0 | — |
case-04 | fail→pass | 9,284 | 2,284 | -75% | 1 | 1 | 0% | 1,769 | 691 | -61% | 0 | 0 | — |
case-05 | pass→pass | 8,512 | 3,846 | -55% | 1 | 1 | 0% | 1,621 | 1,032 | -36% | 0 | 0 | — |
case-06 | pass→pass | 6,859 | 3,307 | -52% | 1 | 1 | 0% | 1,232 | 874 | -29% | 0 | 0 | — |
case-07 | pass→pass | 7,040 | 3,552 | -50% | 1 | 1 | 0% | 1,302 | 940 | -28% | 0 | 0 | — |
case-08 | fail→pass | 3,714 | 2,450 | -34% | 1 | 1 | 0% | 625 | 680 | +9% | 0 | 0 | — |
case-09 | pass→pass | 7,735 | 3,625 | -53% | 1 | 1 | 0% | 1,286 | 971 | -24% | 0 | 0 | — |
case-10 | pass→pass | 2,938 | 1,695 | -42% | 1 | 1 | 0% | 421 | 577 | +37% | 0 | 0 | — |
case-11 | pass→pass | 8,432 | 5,710 | -32% | 1 | 1 | 0% | 1,524 | 1,337 | -12% | 0 | 0 | — |
case-12 | pass→pass | 3,184 | 2,498 | -22% | 1 | 1 | 0% | 416 | 697 | +68% | 0 | 0 | — |
case-13 | fail→fail | 7,290 | 4,834 | -34% | 1 | 1 | 0% | 1,319 | 1,208 | -8% | 0 | 0 | — |
case-14 | pass→pass | 8,355 | 3,277 | -61% | 1 | 1 | 0% | 1,538 | 836 | -46% | 0 | 0 | — |
case-15 | pass→pass | 1,586 | 1,901 | +20% | 1 | 1 | 0% | 231 | 591 | +156% | 0 | 0 | — |
case-16 | fail→pass | 6,955 | 5,283 | -24% | 1 | 1 | 0% | 1,396 | 1,380 | -1% | 0 | 0 | — |
case-17 | pass→pass | 7,537 | 4,642 | -38% | 1 | 1 | 0% | 1,583 | 1,124 | -29% | 0 | 0 | — |
case-18 | fail→pass | 13,631 | 1,921 | -86% | 1 | 1 | 0% | 2,120 | 668 | -68% | 0 | 0 | — |
case-19 | pass→pass | 6,638 | 3,568 | -46% | 1 | 1 | 0% | 1,354 | 956 | -29% | 0 | 0 | — |
case-20 | pass→pass | 7,182 | 4,445 | -38% | 1 | 1 | 0% | 1,325 | 1,086 | -18% | 0 | 0 | — |
case-21 | pass→pass | 4,469 | 2,305 | -48% | 1 | 1 | 0% | 793 | 589 | -26% | 0 | 0 | — |
case-22 | fail→pass | 5,676 | 3,133 | -45% | 1 | 1 | 0% | 887 | 778 | -12% | 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 +27 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.