memory-systems-hardened
Pass
Audited by Gen Agent Trust Hub on Apr 21, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill documentation includes adversarial examples within a safety evaluation report (SAFETY.md) to demonstrate guardrail effectiveness against unauthorized data access and exfiltration attempts.
- [EXTERNAL_DOWNLOADS]: The skill references external tools and repositories for memory management, including Mem0, Zep/Graphiti (github.com/getzep/graphiti), and Cognee (github.com/topoteretes/cognee). These are recognized tools within the AI agent development domain.
- [PROMPT_INJECTION]: The skill facilitates the creation of systems that ingest untrusted data into persistent memory stores, creating a surface for indirect prompt injection. * Ingestion points: Arbitrary data is ingested via cognee.add() and mem0.add() methods in SKILL.md. * Boundary markers: The provided code snippets do not demonstrate the use of delimiters or instructions to ignore embedded commands in stored data. * Capability inventory: The skill enables the agent to perform search and retrieval operations across graph, vector, and relational databases. * Sanitization: No input sanitization or validation logic is included in the implementation examples.
Audit Metadata