multi-ai-research

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

• Third-party content exposure detected (high risk: 0.75). 该 skill 在 Phase 3 通过 opencli grok ask 与 opencli gemini ask 运行外部 AI 查询；这些外部服务会把其检索到的网页/社区内容（公开 web、X 等）以可读文本形式返回到 agent 上下文，从而存在“外部来源自由文本进入 LLM 上下文”的间接提示注入风险。

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill issues runtime Bash calls via opencli to external AI endpoints (e.g., OPENCLI_BROWSER_COMMAND_TIMEOUT=300 opencli grok ask ... → https://grok.com and opencli gemini ask ... → https://gemini.google.com) and then incorporates those AI responses into the agent's arbitration, prompts and action-generation, so these URLs are runtime dependencies that directly control the agent's outputs.