neon-ai-gateway

Pass

Audited by Gen Agent Trust Hub on Jul 3, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill demonstrates how to build agents that process untrusted user input by passing it directly into prompts and message histories for LLMs. This architecture is susceptible to indirect prompt injection, where an attacker could provide malicious instructions that manipulate the agent's behavior or its use of connected tools. \n
  • Ingestion points: User-provided content passed to the prompt and messages parameters in the Vercel AI SDK, Mastra, and OpenAI SDK examples. \n
  • Boundary markers: The examples provided in the documentation do not implement delimiters or specific instructions to isolate or ignore commands within the user-provided data. \n
  • Capability inventory: The skill shows how to define tools that interact with databases, such as the listTodos tool which executes selection queries. \n
  • Sanitization: No input validation or filtering logic is illustrated in the provided integration snippets.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 3, 2026, 01:07 PM
Security Audit — agent-trust-hub — neon-ai-gateway