Agents, tools, and retrieval
The finding is an authority gap. A prompt that only confuses the model is not the report. Use mcp-tool-trust and rag-document-trust for the test.
Tags: llm, mcp, rag, agent
Checklist
- Name the surface: Chat, retrieval, or a tool-calling agent. Write down which one you are allowed to test.
- Tool token: If it calls a tool or an MCP server, open mcp-tool-trust. The question is whether the token is wider than the tool.
- Retrieved text: If it retrieves documents, open rag-document-trust. The question is whether another tenant's text entered a context you are allowed to see.
- Audience: Record the token audience or the chunk boundary you observed. A confusing answer is not that record.
- Output sink: Note where the model text is rendered or stored. The report is the boundary that text crossed.
- Stop: There is no public disclosure card for this shape yet. Do not promote a prompt trick into a finding.