Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Search multiple ML dataset sources (HuggingFace Hub, OpenML, GitHub, Semantic Scholar) and return a ranked, deduplicated list of relevant datasets.
.claude/skills/dataset-discovery/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | — | — |
| case-13 | ✗→✓ | ▲ Improved | — | — |
| case-10 | ✗→✓ | ▲ Improved | — | — |
| case-25 | ✗→✓ | ▲ Improved | — | — |
| case-09 | ✗→✓ | ▲ Improved | — | — |
Search multiple ML dataset sources (HuggingFace Hub, OpenML, GitHub, Semantic Scholar) and return a ranked, deduplicated list of relevant datasets.
Clarify the user's needs before searching:
Run the search script with the user's query:
bashpython3 scripts/search_ml_datasets.py search --query "<query>" --sources huggingface,openml,github,papers --max 30
Options:
--sources: Comma-separated list from huggingface, openml, github, papers. Default: all four.--max: Maximum results to return after dedup + ranking. Default: 30.--modality: Filter by modality (image, text, tabular, audio).--workspace: Output directory. Default: ./datasets/discovery/Optionally also call HF MCP tool hub_repo_search with repo_types: ["dataset"] for semantic search to supplement results.
Show results as a markdown table:
| Name | Source | Downloads | Size | License | Tags | URL | |------|--------|-----------|------|---------|------|-----|
Sort by relevance score (highest first).
When the user wants more info on a specific dataset:
bashpython3 scripts/search_ml_datasets.py detail --dataset-id "huggingface:stanfordnlp/imdb" --workspace ./datasets/discovery/
Writes metadata.json and README.md to {workspace}/datasets/{source}_{slug}/.
When the user wants to preview data:
bashpython3 scripts/search_ml_datasets.py pull --dataset-id "huggingface:stanfordnlp/imdb" --sample-rows 20 --workspace ./datasets/discovery/
Writes sample.jsonl to {workspace}/datasets/{source}_{slug}/.
For full dataset download, confirm with the user first, then use huggingface-cli download or equivalent.
{workspace}/ # default: ./datasets/discovery/
search-{YYYY-MM-DD}.json # search results log
datasets/
{source}_{slug}/
metadata.json # detailed metadata
README.md # human-readable summary
sample.jsonl # sample rowsrequests (stdlib-adjacent, universally available)gh CLI (for GitHub source only)| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-25 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-19 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-24 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-18 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-14 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-02 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-23 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | pass→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
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. 25 cases were attempted, and 22 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +48 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.
Other measured skills in the registry, with their headline benchmark lift.