rag

name	rag
description	Retrieval-augmented generation for accurate, grounded Q&A. Combines search with LLM generation for factual answers with citations.
category	rag
version	1.0.0
tools	["rag_answer","rag_verify","rag_pure_answer","rag_pure_verify"]
triggers	{"keywords":["answer","explain","what is","how does","why","summarize"]}
examples	[{"input":"What is our vacation policy?"},{"input":"Explain the authentication flow"},{"input":"Summarize the Q3 earnings report"}]

RAG Skill

Retrieval-Augmented Generation for accurate, grounded answers from enterprise data.

Available Tools

1. rag_answer - Generate grounded answers with full context

   • Retrieves relevant chunks from indexed documents
   • Generates answers with inline citations
   • Returns confidence scores and grounding metrics

2. rag_verify - Verify claims against source documents

   • Checks if statements are supported by retrieved evidence
   • Returns verification status and supporting passages

3. rag_pure_answer - Minimal RAG answer without tool context

   • Lighter weight for simple questions
   • Faster response time

4. rag_pure_verify - Minimal claim verification

   • Quick fact-checking against sources

Grounding Principles

Every answer must be grounded in source documents:

• ALWAYS cite sources for factual claims
• NEVER generate information not found in documents
• ACKNOWLEDGE when information is partial or uncertain
• DECLINE to answer when sources are insufficient

Confidence Levels

• High (0.8-1.0): Multiple sources confirm the answer
• Medium (0.5-0.8): Single source or partial confirmation
• Low (0.0-0.5): Weak evidence, consider declining to answer

Citation Format

Use inline citations: "The vacation policy allows 20 days PTO [HR Policy Doc, Section 3.2]"

When to Use

• User asks factual questions about company data
• Need verifiable answers with sources
• Summarizing documents or policies
• Explaining technical concepts from documentation