rag-chat

name	rag_chat
description	Chat with AI using relevant documents as context (RAG)
category	ai
version	1.1.0
author	DocSys Team
permissions	["read"]
tags	["ai","rag","retrieval","document-qa","context","knowledge-base"]

RAG Chat

Chat with AI using relevant documents as context (Retrieval-Augmented Generation). Ideal for questions about specific documents or repositories.

Triggers

RAG chat, chat with docs, 基于文档问答, 文档问答, 文档对话, rag, 问文档, ask about docs, question about document, 文档智能问答, 读取文档内容回答, RAG 问答, summarize doc, compare documents

Workflow

1. Parse query: Extract question from user input. Preserve original language.
2. Select model: If user specifies a model → validate it. For analytical/comparative questions → recommend larger model if available.
3. Call CLI: docsys rag <query> [model]
4. ⚠️ Answer quality checkpoint: Inspect the AI answer before presenting to user.
   • If citations present: Present answer with cited document IDs [DocID]. ✓ Safe to show.
   • If no citations: Wrap in ⚠️ warning: "⚠️ No document citations found. Answer may be based on general knowledge or hallucination. Verify key claims against source documents before relying on this answer."
   • If answer is vague/contradictory: Add "⚠️ Answer confidence is low. Consider rephrasing the question or searching with different keywords."
5. No-results fallback: If "No relevant documents found" → suggest broadening query or falling back to docsys search <keywords>.
6. Format output: Present answer with source citations in [DocID] format. Never omit citations warning.

CLI Command

docsys rag <query> [model]
docsys ask <query> [model]

Parameters

Parameter	Type	Required	Description
query	string	yes	Question about documents
model	string	no	AI model to use (default: configured model)

Examples

Example 1

User: "what does the Q4 report say about revenue?" Skill triggers → calls: docsys rag what does the Q4 report say about revenue? Output: AI answer with document citations

Example 2

User: "这份文档的主要观点是什么" Skill triggers → calls: docsys rag 这份文档的主要观点是什么 Output: 基于文档内容回答

Example 3

User: "summarize the key points from the architecture doc" Skill triggers → calls: docsys rag summarize the key points from the architecture doc Output: AI response with citations

Example 4 (vid-filtered)

User: "what is the deployment policy in repo 2" Skill triggers → Step 1: query="deployment policy" → Step 2: vid=2 → calls: docsys rag deployment policy 2 Output: AI answer with citations from documents in repo 2

Output Format

Field	Type	Description
answer	string	AI response text
citations	string[]	Array of cited DocIDs, e.g. [doc_123]
model	string	Model used (if specified)
lang	string	Response language matches input (zh/en)

Success with citations (EN): {answer}\n📎 Sources: {citations.join(", ")} Success with citations (中文): {answer}\n📎 来源：{citations.join(", ")} ⚠️ No citations (EN): ⚠️ No document citations found. Answer may be based on general knowledge or hallucination. Verify key claims against source documents before relying on this answer. ⚠️ No citations (中文): ⚠️ 未找到文档引用。回答可能基于通用知识或虚构内容。请核实关键结论。 ⚠️ Low confidence (EN): ⚠️ Answer confidence is low. Consider rephrasing the question or using different keywords. ⚠️ Low confidence (中文): ⚠️ 回答置信度较低。请尝试换一种表述方式或使用不同关键词。 No documents found (EN): ⚠️ No relevant documents found in repo {vid}. Try: broadening keywords, removing filters, or run 'docsys search <query> {vid}' first. No documents found (中文): ⚠️ 在仓库 {vid} 中未找到相关文档。请尝试：扩大关键词范围、移除过滤器，或先运行 'docsys search <关键词> {vid}' 进行搜索。 Error (EN): ❌ {error_message} Error (中文): ❌ {错误信息}

Error Handling

Error	Cause
Query is required	Empty query
No relevant documents found	Query doesn't match docs
AI service unavailable	LLM service down
Model not found	Invalid model

See references/ for related skills and API docs.