[Feature Request] Memory poisoning defense for agent handoffs

Problem

Swarm agents that maintain context/memory across handoffs are vulnerable to memory poisoning. If adversarial content enters the conversation history or shared context, it persists across agent handoffs and can influence downstream agents' behavior.

Proposed Solution

OWASP Agent Memory Guard can validate context before handoffs:

from agent_memory_guard import scan_memory

# Validate shared context before handoff
for msg in context_variables.get('history', []):
    result = scan_memory(msg['content'])
    if result.blocked:
        # Remove poisoned content before handoff
        context_variables['history'].remove(msg)

• OWASP Incubator project, 5,400+ downloads
• Detects prompt injection, embedding drift, entropy anomalies, PII leakage