claude-rag-skills

The Ailog RAG Skills suite provides a comprehensive toolkit for developers and AI engineers building Retrieval-Augmented Generation systems. It simplifies the end-to-end lifecycle of RAG development by offering specialized commands for auditing existing codebases, evaluating retrieval quality, optimizing chunking strategies, and bootstrapping new projects. Designed for seamless integration into Claude Code workflows, these skills help identify common anti-patterns like inefficient chunking, poor embedding model selection, or lack of hybrid search, ensuring systems are optimized for accuracy and performance.

• Perform deep-dive RAG audits to detect issues in retrieval pipelines, vector store configurations (e.g., Qdrant, Pinecone, Chroma), and embedding strategies.

• Evaluate RAG performance using industry-standard metrics including Recall@K, Precision@K, Mean Reciprocal Rank (MRR), and NDCG.

• Generate production-grade RAG boilerplate code for frameworks like LangChain, LlamaIndex, or vanilla Python implementations.

• Optimize document ingestion via an expert chunking advisor that recommends specific strategies based on document type, such as code, legal contracts, or technical documentation.

• Benchmark local systems against external APIs or reference test datasets to measure faithfulness, relevance, and hallucination guardrails.

• Users should invoke commands directly within the terminal, such as /rag-audit to scan for misconfigurations or /rag-eval to run automated testing against provided JSON datasets.

• Expected inputs include project file paths, target embedding models, and optional API keys (e.g., Ailog API) for advanced benchmarking features.

• Outputs typically consist of structured markdown reports detailing identified critical issues, actionable remediation steps, and scoring metrics for system improvement.

• Note that while the suite is framework-agnostic, it emphasizes industry best practices like query expansion, reranking, hybrid search (dense and sparse), and proper context window management to prevent overflow and noise.