Feature request: OWASP ASI06 memory poisoning guard validator

The Problem

Guardrails AI is designed to validate LLM outputs. However, the current validator library does not include a dedicated validator for ASI06 — Memory Poisoning, the OWASP Agentic Top 10 threat where a malicious payload in external content is written to an agent's memory and poisons future behavior.

The Request

A dedicated MemoryPoisoningGuard validator that:

1. Scans LLM output before it is written to memory
2. Detects prompt injection patterns, jailbreak attempts, and indirect instruction overrides
3. Optionally blocks or flags suspicious content with a configurable policy (block / warn / log)

This maps perfectly to Guardrails' existing validator architecture.

Reference Implementation

The OWASP Agent Memory Guard project provides a lightweight reference implementation (pip install agent-memory-guard). It is already being discussed and adopted by maintainers of LangGraph, LiteLLM, AutoGen, and other major frameworks, and is part of the UK AI Safety Institute's official inspect_evals benchmark suite.

Happy to provide a prototype validator implementation or a draft PR if helpful.