1.16 A2A Protocols & Multi-Session Memory
Validate Agent-to-Agent (A2A) handoff protocols, benchmark multi-session context compaction recall, and red-team persistent memory against OWASP LLM01/LLM06 poisoning attacks.
1. Multi-Agent Conformance Testing
Section titled “1. Multi-Agent Conformance Testing”As enterprise architectures migrate from single agents to agent swarms (orchestrator-worker, router-expert, sub-agent handoffs), CertiQA provides native conformance testing for Agent-to-Agent protocols. A2AAdapter standardizes message envelopes, task delegations, and execution handoffs, while A2AConformanceValidator verifies strict JSON-RPC 2.0 / REST schemas, state preservation across agent transitions, and latency budgets per hop.
2. Multi-Session Memory & Context Compaction Testing
Section titled “2. Multi-Session Memory & Context Compaction Testing”CertiQA tests how agents preserve context across long-running user journeys through two specialized execution engines:
| Subsystem | File Location | Verification Scope |
|---|---|---|
| MultiSessionMemoryRunner | services/certiqa-agent-runner/src/memory/MultiSessionMemoryRunner.ts | Simulates multi-session user journeys (S1 -> S2 -> … -> Sk), testing state recall, user preference retention, and context degradation across independent sessions. |
| ContextCompactionEvaluator | services/certiqa-agent-runner/src/memory/ContextCompactionEvaluator.ts | Benchmarks memory compaction and summarization algorithms against ground-truth fact retention, calculating token savings vs. semantic recall degradation. |
3. Memory Poisoning Attack Catalog
Section titled “3. Memory Poisoning Attack Catalog”Persistent memory introduces novel attack surfaces mapped directly to OWASP LLM01 (Prompt Injection) and OWASP LLM06 (Excessive Agency). The catalog simulates: