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 /messages endpoint as described in SKILL.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
Risk Level
SAFE
Analyzed
Jun 16, 2026, 03:06 AM
Security Audit — agent-trust-hub — conversations