pkai-rag

name	pkai-rag
description	Retrieve grounded evidence from PersonalKnowledgeAI and answer with concise citations for OpenClaw.
metadata	{"openclaw":{"emoji":"📚","always":true}}

PKAI RAG

Use this skill when the user asks a conceptual, explanatory, reflective, or evidence-sensitive question that should first consult the PersonalKnowledgeAI corpus.

Best-fit scenarios

• “什么是……”
• “本质是什么……”
• “为什么会这样……”
• “怎么理解……”
• 需要引用知识库证据的成长、认知、策略、行为类问题

Skip this skill when

• 只是闲聊
• 纯工程排错，且和知识库内容无关
• 用户明确不需要引用或检索

Run

bash {baseDir}/scripts/pkai_rag.sh --query "<user question>" --max-citations 2

Optional tuning:

bash {baseDir}/scripts/pkai_rag.sh --query "<user question>" --top-k 6 --alpha 0.45 --max-citations 2

Output contract

The script returns JSON with:

• answer_hint
• citations
• retrieval_mode
• confidence
• standalone_query

Each citation includes title, section, source, score, snippet, and a compact quote.

Response policy

1. Treat RAG as grounding, not as a replacement for the assistant's reasoning.
2. Prefer 1-2 strong citations over many weak ones.
3. Keep quotes short and evidence-like, not long copy-paste blocks.
4. If no strong hit exists, answer normally and say no direct KB evidence was found.
5. Never fabricate citations.