ai-core/client-persistence
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill manages the persistence of chat messages and generation data originating from external server connections, which constitutes a potential surface for indirect prompt injection.
- Ingestion points: Data is ingested through
useChatand generation hooks via server-sent events or hydration GET requests. - Boundary markers: There are no explicit instructions for prompt delimitation within the persistence logic itself.
- Capability inventory: The skill is designed for UI state management and uses browser storage APIs; it does not request high-risk capabilities like filesystem access or shell command execution.
- Sanitization: Sanitization is not explicitly implemented in the persistence logic, as it is primarily a data-transport and storage concern.
- [DATA_EXFILTRATION]: The server-side hydration endpoint example uses a placeholder
authorizefunction that returnstrue. The documentation includes a necessary warning that developers must implement session-based authorization to prevent unauthorized retrieval of private chat records.
Audit Metadata