line-voice-agent

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The skill documents and uses a built-in web_search tool that "Search[es] the web for real-time information" (SKILL.md Built-in Tools and multiple examples showing tools=[web_search], with fallback to DuckDuckGo), which causes the agent to fetch and incorporate untrusted public web content into LLM reasoning and tool-driven decisions.