What is the main benefit of Contextual Retrieval?

It solves the 'lost context' problem in RAG by ensuring every text chunk contains enough situational information to be correctly identified by search algorithms, even if the query uses document-wide keywords.

When should I avoid using Contextual Retrieval?

Avoid it if your chunks are already self-contained (like Q&A pairs) or if you require real-time, high-volume streaming indexing where the extra LLM step would introduce too much latency.

Is this compatible with Claude 3.5 Sonnet?

Yes, the skill is optimized for Claude 3.5 Sonnet to generate high-quality situational context efficiently and cost-effectively.

How does this skill handle indexing costs?

It implements Anthropic's prompt caching, which caches the document content during the context generation phase, reducing LLM token costs by approximately 90% for large documents.

What is hybrid search in this context?

Hybrid search combines BM25 (keyword matching) and Vector (semantic similarity) scores, which research shows provides superior retrieval performance compared to using either method alone.

Contextual Retrieval RAG

Name: Contextual Retrieval RAG
Author: yonatangross

byyonatangross

•

数据科学与机器学习

Enhances RAG pipelines by prepending situational context to document chunks to preserve semantic meaning and significantly improve retrieval accuracy.

Contextual Retrieval is a production-ready implementation of Anthropic's breakthrough technique for optimizing Retrieval-Augmented Generation (RAG). By prepending succinct, document-level context to individual text chunks before embedding, it solves the common problem where isolated snippets lose critical meaning, such as the document's subject or date. This skill provides a complete pipeline including hybrid search (combining BM25 and Vector scores), parallel batch processing, and prompt caching to reduce indexing costs by up to 90%. It is ideal for developers building high-precision AI applications where retrieval failure is a critical bottleneck.

主要功能

01Async parallel processing for high-performance document ingestion

0269 GitHub stars

03Hybrid search implementation using BM25 and Vector similarity weights

04Cost-optimized indexing via Anthropic's ephemeral prompt caching

05Proven 67% reduction in retrieval failure rates for complex datasets

06Contextualized chunking to preserve document-level metadata and meaning

使用场景

01Optimizing customer support bots that pull information from large knowledge bases

02Enhancing technical documentation search where context like versions or APIs is vital

03Building high-precision search for corporate earnings reports and legal documents

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add yonatangross/orchestkit contextual-retrieval

For use in Claude.ai and ChatGPT

Download Skill