rag-skills

Rag-skills

agent routing RAG skills markdown

A modular collection of best-practice guides and skill definitions for building Retrieval-Augmented Generation (RAG) systems. Designed for AI coding agents, agent frameworks, and teams that want a structured way to route RAG work to the right strategy.

Overview

RAG-skills consolidates actionable skills that help AI agents and builders improve RAG performance, choose appropriate vector databases, implement effective chunking strategies, optimize retrieval quality, and orchestrate multi-step RAG workflows.

Installation

Claude Code

Add this repository as a Claude Code plugin marketplace:

/plugin marketplace add Goodnight77/rag-skills

Then install the RAG skills plugin:

/plugin install rag-skills@rag-skills

Restart Claude Code after installation.

Skills CLI

Install with the Skills CLI:

npx skills add Goodnight77/rag-skills

This installs the root SKILL.md plus the native skill tree under skills/. Claude Code can discover category skills such as /chunking and specific skills such as /semantic-chunking.

Manual Usage

You can also clone the repository and reference the Markdown skills directly:

git clone https://github.com/Goodnight77/rag-skills.git

Then point your agent or coding assistant to the skills/ directory.

Note: This repository follows the Claude Code/Qdrant-style structure: category routers live at paths like skills/chunking/SKILL.md, and specific skills live at paths like skills/chunking/semantic-chunking/SKILL.md.

Skills by Decision Area

This repo is organized as a routing layer for RAG work. Agents can use the category and metadata in each skill file to decide which path to follow for a given problem, instead of treating the repo like a generic reference manual.

Chunking

Use these when the main problem is how to split source material into retrievable units.

• Semantic Chunking - Chunk documents based on semantic boundaries
• Hierarchical Chunking - Multi-level chunking for nested structures
• Sliding Window Chunking - Overlap-based chunking for context preservation
• Contextual Chunk Headers - Adding higher-level context to chunks

Vector Databases

Use these when the main problem is choosing or operating the storage layer for embeddings and metadata.

• Qdrant Setup for RAG - Setting up Qdrant for RAG
• Qdrant for Production RAG - Scaling RAG with Qdrant
• Choosing Vector DB by Datatype - Database selection guide

Retrieval Strategies

Use these when the main problem is search quality, ranking, recall, or combining search methods.

• Hybrid Search BM25 Dense - Combining keyword and semantic search
• Multi-Pass Retrieval with Reranking - Two-pass retrieval with cross-encoder reranking
• Query Transformation Strategies - Query rewriting, step-back prompting, sub-query decomposition
• HyDE - Hypothetical Document Embeddings - Query expansion with LLM-generated documents
• HyPE - Hypothetical Prompt Embeddings - Precomputed question embeddings at indexing time
• Self-RAG - Self-reflective retrieval with relevance evaluation
• RAPTOR - Hierarchical Retrieval - Multi-level tree of document summaries
• Context Enrichment Window - Adding surrounding chunks to retrieved results
• Adaptive Retrieval - Dynamic strategy selection based on query type
• Explainable Retrieval with Citations - Traceability and source attribution
• CRAG - Corrective RAG - Dynamic correction with web search
• Graph RAG - Knowledge graph-based retrieval