skills.rest
kridaydave
kridaydaveCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

trove

Builds datasets through sourcing, sampling, labeling, deduplication, splitting, and datasheet documentation.

1|Updated Jun 1, 2026
Search Tags:#data collection#data leakage#datasheet#sampling strategy#data labeling#dataset construction#train test split
One-click install
npx skills add https://github.com/kridaydave/My_Skills --skill trove-kridaydave
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: trove
Source: https://github.com/kridaydave/My_Skills/tree/main/trove
Command: npx skills add https://github.com/kridaydave/My_Skills --skill trove-kridaydave

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Datasets built carelessly cause the most expensive failures in data work: train/test leakage, sampling bias, label noise, and licensing violations surface only after models fail in production. This Skill guides the entire dataset construction stage before any analysis or training begins. ## Core Features & Use Cases - Source & License Vetting: Plans data collection from APIs, scraping, or existing corpora with provenance, license, ToS, and consent checks before anything is pulled. - Sampling & Labeling Design: Defines target populations, sampling strategies, class balance decisions, operational label schemes, and inter-annotator agreement measurement (κ/α). - Leakage-Proof Splits & Datasheets: Deduplicates before splitting, splits by time/entity/group boundaries, and produces a datasheet documenting biases, gaps, and intended use. - Use Case: You need training data for a product-review sentiment classifier. The Skill rejects a generic movie-review dataset, plans stratified sampling from your own reviews, defines a 3-class labeling scheme with agreement targets, and builds leakage-free splits with a full datasheet. ## Quick Start Ask the assistant to help you build a dataset for your task, including where to source the data, how to sample and label it, and how to split it without leakage.

Frequently Asked Questions about trove

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I build a dataset for training a machine learning model?▼

Start from the downstream claim: define the target population and unit, plan sourcing with license checks, design a sampling strategy covering rare classes, then clean, label, split, and document. The Skill outputs a full dataset spec covering each stage.

How do I prevent train/test leakage when splitting data?▼

Split by the boundary the real world enforces: by time for forecasting, by entity or group when rows cluster (e.g., customer_id). Deduplicate before splitting so no record straddles train and test, and state the split key used.

Can I scrape websites to build a training dataset?▼

Not automatically. Public availability does not equal legal usability — check the site's terms of service, copyright, and consent requirements first. The Skill flags restrictive sources and proposes licensed or open alternatives.

How do I design a labeling scheme that annotators apply consistently?▼

Define each label operationally with explicit criteria and worked examples, then double-label a subset and measure inter-annotator agreement (κ or α). Below the threshold, fix the guidelines rather than blaming annotators, and adjudicate ties by a written rule.

How many labeled examples do I need for a classifier?▼

It depends on the downstream need, not a round number — the minority class is the binding constraint, not the total. Run a learning-curve pilot: label in representative batches, plot performance against n, and stop when the curve flattens.

What should a dataset datasheet include?▼

A datasheet records what the dataset is, how it was built, known biases and coverage gaps, what it is not for, and its license. It makes cleaning choices, sampling decisions, and provenance auditable for anyone using the data later.

Related Skills

Explore More Data & Analytics Skills→
adobeadobe
1.5k

ask-questions-if-underspecified

Identify underspecified requirements and prompt 1-5 must-have questions before implementation.

Official
Intermediate
CRAG666CRAG666
46

ml-science-discipline

Enforce rigorous ML experimental design with leakage controls and publication-standard reporting.

Community
Advanced
micpossomicposso

privacy

Analyze AI/LLM applications for biometric, PII, and surveillance privacy violations.

Community
Intermediate
rappdwrappdw

synthdata-extend

Extend synthetic datasets with rows or columns while preserving FK integrity and ID sequencing.

Community
Intermediate