| name | query-retriever |
| description | Forward the user's question VERBATIM to the data retriever agent - never rephrase or add assumptions. Use for MCP, MCP tools or resources, data retrieval, documents, questions about employees, enterprise data, authorization (AuthZEN), knowledge queries, or any internal/knowledge-graph data. The retriever uses an MCP server with access to enterprise data. |
| tags | ["retriever","mcp","data","enterprise","authzen"] |
| examples | ["What data do we have about employee X?","List available MCP resources.","Can Alice view record 109?","Search documents about project Y.","What can James Mitchell see about the home insurance?"] |
Query Retriever
Use this skill when the user asks for data that the retriever agent can provide: MCP resources or tools, documents, employees, enterprise data, AuthZEN authorization, knowledge-graph queries, or any internal data.
Tool
- query_retriever – Sends the user's question to the data retriever agent via A2A and returns the retriever's response. Pass the question as the
query argument.
Passing the question
Pass the user's question verbatim as query - do not rephrase, expand,
or add assumptions (e.g. never turn "our car" into "our company car"; a
logged-in customer's "our" means their household, not the company). The
retriever only sees what you pass. The single exception: if the message
cannot stand alone (pronouns, short follow-ups like "and what about Sarah?"),
add only context the user actually stated earlier in the conversation - never
invented detail.
When to use
- User asks about enterprise data, documents, employees, MCP resources, authorization (e.g. "Can X do Y?"), household/policy data, or any topic the retriever handles.
- Prefer query_retriever over web search for data, documents, employees, and authorization. Only use web search when the information is real-time or external and cannot be answered from enterprise data.
