What is provenance tracking in this context?

Provenance tracking is the architectural design that ensures every piece of information used by the LLM is linked to its source metadata, such as document IDs, graph nodes, or confidence scores, for transparency.

Can I use this for non-graph RAG systems?

While specifically optimized for Knowledge Graphs (GraphRAG), the underlying logic for search orchestration, ranking constraints, and provenance tracking is applicable to many high-end RAG architectures.

What is the benefit of U-Shaped retrieval?

U-Shaped retrieval is a hybrid approach that combines broad thematic coverage (Global-First) with specific entity-centric details (Local-First) to provide the most comprehensive context for complex queries.

Does it support structured query generation?

Yes, it includes patterns for tool-augmented retrieval where Claude can generate formal database queries like Cypher or SPARQL to perform structured graph traversals.

How does this skill handle multi-hop reasoning?

It utilizes query decomposition patterns like ReAct, Self-Ask, and LLM-as-controller to break complex, multi-layered questions into smaller sub-queries that can be resolved sequentially or in parallel.

Retrieval & Search Orchestration

This skill empowers developers to design and implement advanced retrieval architectures for GraphRAG and standard RAG systems. It provides structured guidance on selecting the optimal retrieval pattern—such as Global-First, Local-First, or U-Shaped Hybrid—while offering robust strategies for query decomposition using LLM-as-controller or ReAct patterns. By focusing on provenance tracking and evidence-based generation, it ensures that every answer is backed by verifiable source data, making it indispensable for complex domain-specific AI applications like biomedical research, legal analysis, or technical documentation search.

主な機能

01Configurable ranking constraints and similarity thresholds for high-precision results

02Comprehensive provenance tracking for verifiable evidence-based answers

03Advanced query decomposition including ReAct and Self-Ask patterns

04Iterative fallback strategies for handling failed or insufficient retrieval attempts

05Pattern selection guide for Global, Local, and U-Shaped retrieval strategies

0632 GitHub stars

ユースケース

01Building complex GraphRAG systems that require multi-hop reasoning across interconnected data

02Optimizing retrieval performance for large-scale knowledge graphs with diverse query types

03Developing enterprise search tools where answer provenance and source citations are critical requirements

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add lyndonkl/claude retrieval-search-orchestration

For use in Claude.ai and ChatGPT

Download Skill