tanstack-ai-memory
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents a pattern where data retrieved from external memory adapters (like Redis, Mem0, or Honcho) is interpolated directly into the system prompt via the
recallmethod. - Ingestion points: Data enters the agent context through
adapter.recallinside thememoryMiddlewareas shown inSKILL.md. - Boundary markers: The provided examples do not explicitly define delimiters or instructions for the model to ignore malicious content embedded within retrieved memories.
- Capability inventory: The retrieved memory content influences the
chat()call which determines the model's response and tool usage. - Sanitization: No explicit sanitization or validation of the retrieved memory text is mentioned before it is injected into the system prompt.
- [SAFE]: The documentation includes explicit security guidance regarding scope isolation, warning developers to never trust client-supplied identifiers and to resolve sessions server-side, which aligns with standard security best practices.
Audit Metadata