メニュー
Expert guidance on RAG (Retrieval-Augmented Generation) system design including chunking strategies, embedding selection, retrieval methods, and vector database choices
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Automated visual testing with Playwright MCP - test web apps, presentations, websites, and documents with scalable reviewer perspectives
Design complex system architectures, evaluate tradeoffs, and make critical technical decisions requiring deep reasoning
Deep code analysis identifying subtle bugs, security issues, performance problems, and architectural concerns requiring expert-level reasoning
Efficient context state inspection, task lifecycle management, and session tracking
Systematic codebase onboarding. Builds a mental model of a new or unfamiliar project by exploring structure, architecture, key data models, entry points, and auth patterns.
Systematically extract knowledge from scattered documentation to prevent catastrophic forgetting. Creates structured extraction reports with status tags.
| name | ai-rag-advisor |
| description | Expert guidance on RAG (Retrieval-Augmented Generation) system design including chunking strategies, embedding selection, retrieval methods, and vector database choices |
Purpose: Expert guidance on RAG (Retrieval-Augmented Generation) system design including chunking strategies, embedding selection, retrieval methods, and vector database choices.
Model: Sonnet 4.5 (balanced reasoning for RAG architecture, up-to-date on 2026 techniques)
When providing RAG guidance, you can reference this workflow's working implementation:
Architecture: GraphRAG (semantic search + graph traversal)
[[uses::Tool]], [[implements::Concept]]Vector Database: Weaviate
ClaudeKnowledgeGraph, [Project]_developmentEmbeddings: snowflake-arctic-embed2
Chunking Strategy:
Search Methods:
.claude/scripts/kg-search (~100ms) for exact termshybrid_search() (~500ms) for conceptssemantic_graph_search() (~1-2s) for relationshipshybrid_search() (~1-2s) combines all threeStorage Pattern:
Collections Strategy:
ClaudeKnowledgeGraph: Concise cross-project patterns (<300 lines/node)[Project]_development: Verbose project-specific docs (no size limit)
Performance:This is a working implementation that balances performance, cost (free local embeddings), and quality. Use it as reference when advising on RAG systems.
Provides guidance on chunking methods:
Recommendations across categories:
Analyzes retrieval approaches:
Compares databases:
Outlines patterns:
Performance improvements:
See template.md for complete RAG system design structure.
After RAG design:
knowledge/concepts/rag-[use-case]-strategy.md