agent-memory-systems
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill demonstrates secure handling of sensitive information by instructing the use of environment variables for database connection strings and API keys (e.g., POSTGRES_URL, PINECONE_API_KEY).
- [PROMPT_INJECTION]: Surface for indirect prompt injection detected in memory ingestion and retrieval workflows. * Ingestion points: User conversation data is ingested and processed in the LangGraph background memory processor (SKILL.md). * Boundary markers: Instruction templates for memory extraction do not use explicit delimiters or 'ignore embedded instructions' warnings. * Capability inventory: The skill utilizes various vector database clients and LangMem to perform data persistence and search operations. * Sanitization: No explicit sanitization, validation, or filtering of retrieved memory content is demonstrated before it is re-interpolated into the agent's context.
- [EXTERNAL_DOWNLOADS]: The skill refers to well-known and established technology services and libraries such as LangChain, Pinecone, Qdrant, and Weaviate for its implementation examples.
Audit Metadata