conversation-memory
Pass
Audited by Gen Agent Trust Hub on Aug 9, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill logic for entity extraction presents an indirect prompt injection vulnerability. Untrusted user input is directly included in prompts used for memory storage operations.\n
- Ingestion points: The
extractAndStoreandaddMessagemethods inSKILL.mdingest untrusted content from themessage.contentfield.\n - Boundary markers: The prompt template used for entity extraction lacks delimiters or instructions to the LLM to treat interpolated content as data rather than instructions.\n
- Capability inventory: The code demonstrates persistent storage, search, and structured output parsing, but does not include high-risk capabilities like shell access or network requests to untrusted domains.\n
- Sanitization: No validation or escaping is applied to the user-supplied content before it is stored or used in prompt generation.
Audit Metadata