rag-boundary-check

name	rag-boundary-check
description	Verify the real RAG boundaries in this hospital project. Use to confirm knowledge sources, retrieval mode, whether it is explanation-only, and what is intentionally not handled by RAG.

RAG Boundary Check

Use this skill when you need a code-accurate answer about what the current RAG implementation does and does not do in this repository.

What this skill is for

Answer questions like:

• Is the current RAG a vector database solution?
• Does RAG handle real-time registration facts?
• Is RAG only used for explanations?
• Is retrieval keyword-only, vector-only, or hybrid?
• What documents are used as the knowledge source?

Scope

Focus on:

• patient-wx-api-mysql/src/main/java/com/example/hospital/patient/wx/api/agent/multi/rag/
• patient-wx-api-mysql/src/main/java/com/example/hospital/patient/wx/api/agent/multi/service/MultiAgentCoordinatorService.java
• docs/agent/multi-agent.md

Required reading order

1. docs/agent/multi-agent.md
2. patient-wx-api-mysql/src/main/java/com/example/hospital/patient/wx/api/agent/multi/service/MultiAgentCoordinatorService.java
3. patient-wx-api-mysql/src/main/java/com/example/hospital/patient/wx/api/agent/multi/rag/MultiAgentRagService.java
4. patient-wx-api-mysql/src/main/java/com/example/hospital/patient/wx/api/agent/multi/rag/MultiAgentKnowledgeBase.java
5. patient-wx-api-mysql/src/main/java/com/example/hospital/patient/wx/api/agent/multi/rag/EmbeddingClient.java
6. patient-wx-api-mysql/src/main/java/com/example/hospital/patient/wx/api/agent/multi/rag/EmbeddingVectorIndex.java

Output rules

• Distinguish clearly between static explanation knowledge and real-time business facts.
• Explicitly state whether the implementation uses:
  • markdown slicing
  • keyword retrieval
  • embedding retrieval
  • hybrid retrieval
  • independent vector database or not
• State whether RAG participates in write decisions.
• Use file_path:line_number anchors.

Important project-specific conclusions to verify, not assume

• Current RAG is explanation-oriented, not the source of live schedule or registration truth.
• Knowledge comes from docs/agent/*.md, not a hospital-wide FAQ platform.
• Retrieval is lightweight hybrid retrieval when embedding is enabled.
• The current implementation uses in-process indexing/vector cache rather than a standalone vector database.
• Safe memory is deliberately trimmed before being sent into the RAG generation path.

Good response shape

1. One-sentence conclusion
2. What RAG is used for
3. What RAG is not used for
4. Retrieval/data source details
5. Key code anchors