Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Generate structured datasheets for datasets (Gebru et al. "Datasheets for Datasets" format) — purpose, composition, collection process, preprocessing, limitations, and licensing.
.claude/skills/mkurman-data-card-writer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-12 | ✗→✓ | ▲ Improved | 34% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -8% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -8% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 29% | 0% |
Generates a structured datasheet following the "Datasheets for Datasets" framework (Gebru et al., 2021). Every dataset should ship with a data card that answers: who made this, what's in it, how was it collected, what are the limitations.
markdown# [Dataset Name] — Data Card v[Version] ## Motivation - Purpose: [What task is this for?] - Creator: [Who created it?] - Funding: [Who funded it?] ## Composition - Instances: [N] examples, [M] features - Target: [What is being predicted?] - Protected attributes: [Gender, race, age, etc. — or "not collected"] - Missing data: [X% missing overall, per-column breakdown] - Class balance: [Distribution] ## Collection Process - Source: [URL, database, API, instrument] - Collection date: [YYYY-MM-DD to YYYY-MM-DD] - Sampling strategy: [Random, stratified, convenience] - Ethical review: [IRB protocol # or "not applicable"] - Consent: [How was consent obtained?] ## Preprocessing - Raw → cleaned pipeline: [Steps applied] - Exclusions: [What was removed and why?] - Transformations: [Normalization, imputation, encoding] - Cleaning script hash: [sha256] ## Uses - Recommended uses: [What this dataset is validated for] - Discouraged uses: [What this should NOT be used for] - Out-of-scope: [Inappropriate applications] ## Distribution - License: [e.g., CC-BY-4.0, CDLA-Permissive] - Access: [URL or process] - Version: [Semantic version] ## Limitations - Known biases: [Demographic, temporal, geographic] - Coverage gaps: [Missing populations, conditions, domains] - Label quality: [Inter-rater agreement, noise estimates] ## Maintenance - Maintainer: [Contact or organization] - Update frequency: [Monthly, annually, none] - Errata: [Link to corrections] ## Citation [BibTeX or DOI]
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-12 | fail→pass | 15,219 | 18,554 | +22% | 1 | 1 | 0% | 2,770 | 3,725 | +34% | 0 | 0 | — |
case-04 | pass→pass | 17,854 | 15,943 | -11% | 1 | 1 | 0% | 3,214 | 3,561 | +11% | 0 | 0 | — |
case-11 | pass→pass | 13,490 | 17,662 | +31% | 1 | 1 | 0% | 2,219 | 3,774 | +70% | 0 | 0 | — |
case-01 | fail→pass | 18,432 | 13,166 | -29% | 1 | 1 | 0% | 3,156 | 2,901 | -8% | 0 | 0 | — |
case-02 | fail→pass | 17,158 | 13,458 | -22% | 1 | 1 | 0% | 3,114 | 3,048 | -2% | 0 | 0 | — |
case-03 | fail→pass | 15,048 | 12,147 | -19% | 1 | 1 | 0% | 2,872 | 2,630 | -8% | 0 | 0 | — |
case-05 | pass→pass | 16,335 | 14,556 | -11% | 1 | 1 | 0% | 3,075 | 3,326 | +8% | 0 | 0 | — |
case-06 | pass→fail | 19,518 | 21,199 | +9% | 1 | 1 | 0% | 3,914 | 4,754 | +21% | 0 | 0 | — |
case-07 | pass→pass | 20,455 | 15,854 | -22% | 1 | 1 | 0% | 3,432 | 3,333 | -3% | 0 | 0 | — |
case-08 | fail→pass | 17,620 | 18,749 | +6% | 1 | 1 | 0% | 3,104 | 3,990 | +29% | 0 | 0 | — |
case-09 | pass→pass | 23,237 | 17,438 | -25% | 1 | 1 | 0% | 4,448 | 3,833 | -14% | 0 | 0 | — |
case-10 | fail→pass | 17,876 | 17,345 | -3% | 1 | 1 | 0% | 3,010 | 3,630 | +21% | 0 | 0 | — |
case-13 | fail→pass | 17,367 | 17,985 | +4% | 1 | 1 | 0% | 2,677 | 3,624 | +35% | 0 | 0 | — |
case-14 | pass→pass | 19,491 | 13,515 | -31% | 1 | 1 | 0% | 3,334 | 2,787 | -16% | 0 | 0 | — |
case-15 | fail→pass | 22,853 | 20,401 | -11% | 1 | 1 | 0% | 3,532 | 4,199 | +19% | 0 | 0 | — |
case-16 | fail→pass | 20,570 | 17,298 | -16% | 1 | 1 | 0% | 3,369 | 3,545 | +5% | 0 | 0 | — |
case-17 | fail→pass | 24,411 | 19,973 | -18% | 1 | 1 | 0% | 4,203 | 4,073 | -3% | 0 | 0 | — |
case-18 | pass→pass | 18,375 | 12,965 | -29% | 1 | 1 | 0% | 3,033 | 2,796 | -8% | 0 | 0 | — |
case-19 | fail→pass | 14,585 | 15,093 | +3% | 1 | 1 | 0% | 2,465 | 3,411 | +38% | 0 | 0 | — |
case-20 | fail→pass | 17,857 | 21,901 | +23% | 1 | 1 | 0% | 3,012 | 4,305 | +43% | 0 | 0 | — |
case-21 | fail→pass | 19,382 | 13,708 | -29% | 1 | 1 | 0% | 3,319 | 2,946 | -11% | 0 | 0 | — |
case-22 | pass→fail | 15,104 | 17,477 | +16% | 1 | 1 | 0% | 2,678 | 3,765 | +41% | 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 +50 percentage points is the difference between those two pass rates over the 22 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.
Other measured skills in the registry, with their headline benchmark lift.