conversations
Pass
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill provides an interface to retrieve external message history, introducing a vulnerability to indirect prompt injection. \n
- Ingestion points: Untrusted message content from users or other models is retrieved via the
/messagesendpoint as described inSKILL.md. \n - Boundary markers: No instructions or delimiters are present to help the agent distinguish data from its primary instructions. \n
- Capability inventory: The skill possesses potentially destructive capabilities, including the ability to delete conversations (
DELETE /conversations/{id}) and modify session variables. \n - Sanitization: The skill description does not specify any sanitization or validation of the ingested conversation data.\n- [SAFE]: The skill uses standard HTTP requests to the official Dify API domain (
api.dify.ai) and employs standard security practices such as placeholder tokens for authentication.
Audit Metadata