What is a golden dataset in the context of Claude Code?

A golden dataset is a curated collection of high-quality 'ground truth' data used to benchmark and evaluate the performance of AI models and RAG systems.

Does it handle duplicate content?

Yes, the skill includes a duplicate check that prevents adding content with more than 80% similarity to existing entries in your dataset.

What content types are supported for extraction?

The skill can detect and extract structure from blog articles, step-by-step tutorials, API documentation, and academic research papers.

How does the quality scoring system function?

It uses a weighted scoring system across four dimensions: Accuracy (0.25), Coherence (0.20), Depth (0.25), and Relevance (0.30). Documents scoring above 0.75 are included automatically, while those between 0.55 and 0.75 require review.

How does the multi-agent validation work?

The skill employs four parallel agents: one for code quality review and three 'Explore' agents that handle difficulty classification, domain tagging, and test query generation.

Golden Dataset Curator

Name: Golden Dataset Curator
Author: yonatangross

byyonatangross

•

セキュリティとテスト

Curates and validates high-quality documents for golden datasets using a multi-agent evaluation workflow.

概要

The add-golden skill streamlines the creation of high-quality evaluation datasets for LLMs and RAG systems by automating the curation process. Using a sophisticated multi-agent architecture, it fetches web content, classifies document types, and performs parallel analysis to score quality across dimensions like accuracy and relevance. It ensures dataset integrity through automated duplicate detection, schema validation, and difficulty classification, providing developers with a robust, production-ready pipeline for managing ground-truth data within Claude Code.

主な機能

Multi-agent validation with four specialized analysis agents
Automatic content extraction and difficulty classification
Quality-gate scoring based on accuracy, coherence, and depth
Seamless integration with JSON fixture files for evaluation
29 GitHub stars
Automated duplicate detection and 80% similarity checking

ユースケース

Curating high-quality training examples from technical documentation
Automating the expansion of test suites with verified web content
Building ground-truth datasets for RAG system evaluation

概要

主な機能

Multi-agent validation with four specialized analysis agents
Automatic content extraction and difficulty classification
Quality-gate scoring based on accuracy, coherence, and depth
Seamless integration with JSON fixture files for evaluation
29 GitHub stars
Automated duplicate detection and 80% similarity checking

ユースケース

Curating high-quality training examples from technical documentation
Automating the expansion of test suites with verified web content
Building ground-truth datasets for RAG system evaluation