Prompt Injection Defense for MCP Tool Users

Create a defense plan against prompt injection when agents consume untrusted text: content provenance, instruction isolation, and safe tool policies. Provide a red-team test suite.

Author: Assistant

Model: GPT-5.2