RAG Chunking Strategy Advisor

RAG Chunking Strategy Advisor

What this skill does

This skill analyzes your document types, content structure, and retrieval goals to recommend the right chunking strategy for your RAG pipeline. Poor chunking is the #1 cause of RAG failures — chunks too large lose precision, chunks too small lose context. This skill picks the right strategy and explains exactly how to implement it.

How to use

Claude Code / Cline

Copy this file to .agents/skills/rag-chunking-advisor/SKILL.md in your project root.

Then ask:

• "Use the RAG Chunking Strategy Advisor to help me chunk our legal contract PDFs."
• "What chunking strategy should I use for markdown documentation with code blocks?"

Provide:

• Document type (PDFs, markdown, HTML, code, emails, etc.)
• Typical document length
• What users will search for (questions, keywords, concepts)
• Your embedding model if known

Cursor / Codex

Paste the instructions below along with your document type and retrieval use case.

The Prompt / Instructions for the Agent

When asked to advise on RAG chunking, follow these steps:

Step 1 — Identify document characteristics

Ask or infer:

• Document type: structured (tables, headers) vs. unstructured (prose) vs. code
• Length: short ( "Use the RAG Chunking Advisor. I'm building a support chatbot. Documents are markdown product docs, 5–30 pages each, with H2 and H3 headers. Users ask specific how-to questions."

Output: > Recommended strategy: Header-based chunking > > Split on H2 headers as primary boundaries, H3 as secondary. Each chunk = one section, with its full heading path prepended (Product > Installation > macOS Setup). > > Config: > ``python > from langchain.text_splitter import MarkdownHeaderTextSplitter > splitter = MarkdownHeaderTextSplitter( > headers_to_split_on=[("##", "section"), ("###", "subsection")] > ) > ` > > Add metadata: {"source": filename, "section": headingpath, "doctype": "productdocs"} > > If sections exceed 800 tokens, apply a secondary recursive split with chunksize=600, overlap=60`. > > Validate: run your top 20 support questions against the index and confirm the relevant section appears in top-3.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Notysoty
• Source: Notysoty/openagentskills
• License: MIT

Install and usage instructions live in the source repository linked above.