Add instructions for Nova to query RAG directly

The memory_search tool requires external API keys (OpenAI/Google/Voyage)
which are not configured. This documents that Nova should use the RAG
scripts directly via command line instead.

CHANGELOG.md

@@ -2,6 +2,20 @@
 
 All notable changes to the OpenClaw RAG Knowledge System will be documented in this file.
 
+## [1.0.7] - 2026-02-14
+
+### Fixed
+- **Bug in chunk_messages()**: Fixed undefined variable `session_key` referenced in metadata generation
+  - Added `session_key` parameter to `chunk_messages()` function signature
+  - Fixed bug identified in ClawHub security scan report
+  - Pass `session_key` from ingestion loop to chunk_messages() call
+  - Resolves scope issue where function referenced non-existent variable
+
+### Security
+- Fixes code quality issue identified in security scan (bug in implementation)
+
+---
+
 ## [1.0.6] - 2026-02-14
 
 ### Changed

SKILL.md

@@ -18,6 +18,32 @@ This skill provides a complete RAG (Retrieval-Augmented Generation) system for O
 - 💾 Local ChromaDB storage (no API keys required)
 - 🚀 Automatic AI integration – retrieves context transparently
 
+## For Nova (AI Assistant) - How to Query RAG
+
+**IMPORTANT:** The `memory_search` tool in OpenClaw is a SEPARATE system that requires external API keys (OpenAI/Google/Voyage). Do NOT use it.
+
+**Use the RAG skill scripts directly:**
+
+```bash
+# Quick search (returns top 10 results)
+python3 ~/.openclaw/workspace/skills/rag-openclaw/rag_query.py "your search query"
+
+# Search with type filter
+python3 ~/.openclaw/workspace/skills/rag-openclaw/rag_query.py "SIP voip" --type session
+python3 ~/.openclaw/workspace/skills/rag-openclaw/rag_query.py "porkbun DNS" --type skill
+
+# Check what's indexed
+python3 ~/.openclaw/workspace/skills/rag-openclaw/rag_manage.py stats
+```
+
+**When to use RAG:**
+- Before answering questions about past work, decisions, or context
+- When the user says "remember when..." or "we set up..."
+- When you need to recall configurations, credentials, or setup details
+- Whendebugging - search for how similar issues were solved before
+
+The RAG uses local embeddings (all-MiniLM-L6-v2) and requires no external API keys.
+
 ## Installation
 
 ### Prerequisites

ingest_sessions.py

@@ -127,6 +127,7 @@ def format_content(content) -> str:
 
 def chunk_messages(
     messages: List[Dict],
+    session_key: str,
     context_window: int = 20,
     overlap: int = 5
 ) -> List[Dict]:
@@ -135,6 +136,7 @@ def chunk_messages(
 
     Args:
         messages: List of message objects
+        session_key: Session identifier for metadata
         context_window: Messages per chunk
         overlap: Message overlap between chunks
 
@@ -244,7 +246,7 @@ def ingest_sessions(
         print(f"   Messages: {len(messages)}")
 
         # Chunk messages
-        chunks = chunk_messages(messages, chunk_size, chunk_overlap)
+        chunks = chunk_messages(messages, session_key, chunk_size, chunk_overlap)
 
         if not chunks:
             print(f"   ⚠️  No valid chunks, skipping")

package.json

@@ -1,6 +1,6 @@
 {
   "name": "rag-openclaw",
-  "version": "1.0.6",
+  "version": "1.0.7",
   "description": "RAG Knowledge System for OpenClaw - Semantic search across chat history, code, docs, and skills with automatic memory retrieval",
   "homepage": "https://openclaw-rag-skill.projects.theta42.com",
   "author": {