HallNo RAG has 6 tools with basic schemas and descriptions, but significant gaps prevent production use. Tool names lack clear action verbs (manage_agent_system is vague). Descriptions are present but generic (10-50 chars for most). Parameter descriptions are minimal or absent. No output schemas documented. Error handling is absent. The query tool conflates multiple concerns (RAG + multi-agent orchestration). No pagination, no field-level documentation, no recovery guidance.

• Split manage_agent_system into three separate tools: initialize_agent_system, shutdown_agent_system, get_agent_system_status. Each should have a clear verb and single responsibility.
• Split query into query_rag (retrieves documents from vector store) and query_with_agents (orchestrates multi-agent reasoning). Document output schema for each: query_rag returns [{ document_id, content, relevance_score }]; query_with_agents returns { answer, agent_reasoning, sources }.
• Add parameter descriptions with format constraints: file_extensions should specify 'Array of file extensions without dots (e.g., ["py", "js", "ts"])'. context should document accepted keys (e.g., 'Optional object with keys: domain, max_tokens, temperature').
• Document output schemas for all tools. add_project should return { project_id: string, path: string, file_count: int, status: 'added'|'already_monitored' }. list_projects already has ProjectListResponse but should be declared in tool definition.
• Add error handling with recovery guidance. If add_project fails with 'Path not found', return 'Project path does not exist. Verify the path is absolute and accessible.' If 'Already monitored', return 'Project already being monitored. Call remove_project first to re-add with different extensions.'