|
@@ -0,0 +1,34 @@
|
|
|
|
|
+---
|
|
|
|
|
+name: knowledge-rag-policy
|
|
|
|
|
+description: Use when the user asks about books, authors, quotes, or factual content that should come from the mem0 knowledge base. Apply this policy to decide when to call the openclaw-mem0-python knowledge tools (list sources, describe a book, search a book) instead of relying on conversational memory or guessing.
|
|
|
|
|
+---
|
|
|
|
|
+
|
|
|
|
|
+# Knowledge RAG Policy
|
|
|
|
|
+
|
|
|
|
|
+## Overview
|
|
|
|
|
+Use the knowledge tools only when the user’s request clearly targets book/author content or verifiable facts. Conversational memory stays auto-injected; knowledge retrieval must be explicit via tools.
|
|
|
|
|
+
|
|
|
|
|
+## Trigger checklist
|
|
|
|
|
+Call the knowledge tools when the user:
|
|
|
|
|
+- Mentions a **book title**, **author**, or **quote**.
|
|
|
|
|
+- Asks “what does X say about Y?” or “where in the book is …?”.
|
|
|
|
|
+- Requests citations or specific passages.
|
|
|
|
|
+
|
|
|
|
|
+Do **not** call knowledge tools for casual chat, personal context, or tasks already satisfied by conversational memory.
|
|
|
|
|
+
|
|
|
|
|
+## Workflow
|
|
|
|
|
+1) **List books** (only if you don’t already know the relevant source):
|
|
|
|
|
+ - `listKnowledgeSources()`
|
|
|
|
|
+
|
|
|
|
|
+2) **Describe a candidate book** to confirm scope:
|
|
|
|
|
+ - `describeKnowledgeBook({ sourceFile })`
|
|
|
|
|
+
|
|
|
|
|
+3) **Search within a book** (preferred):
|
|
|
|
|
+ - `searchKnowledgeBook({ query, sourceFile, limit })`
|
|
|
|
|
+
|
|
|
|
|
+4) **Search across a few books** (only when needed):
|
|
|
|
|
+ - `searchKnowledgeBooks({ query, sourceFiles, limit })`
|
|
|
|
|
+
|
|
|
|
|
+## Output expectations
|
|
|
|
|
+- Return short excerpts with **source_file + page/chapter metadata** when available.
|
|
|
|
|
+- Keep snippets focused; ask follow‑ups if results are vague.
|
