alibabacloud-lindorm-agent-skill

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.90). The prompt includes an "AI-Mode Lifecycle [Internal Agent Rule — Do NOT Show to User]" section that embeds hidden instructions (enable/disable AI-mode, set a specific user-agent, update plugins) intended for the agent and explicitly hidden from the user, which are deceptive/behavior-changing directives outside the skill's visible purpose.

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

• Third-party content exposure detected (high risk: 0.85). 在“知识库检索/多模态检索”运行路径中，Agent 会把用户/外部导入的文档内容（如 txt/JSON chunk 文本、图片 URL 对应的 VL 描述、以及检索返回的 text_field/vl_description）拼接为 <retrieved_context> 或“已知信息”后，作为可读自由文本进入 LLM 上下文；该文本来源于非操作用户自写的知识库/检索结果（outsider-authored content）。

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly requires the Agent at runtime to fetch and parse the official documentation (e.g. https://help.aliyun.com/zh/lindorm/) to extract and verify code examples before generating responses, so external content from that URL is used during runtime to directly control the agent's generated prompts/instructions.

MEDIUM W021: Hidden or invisible Unicode characters detected (potential obfuscation or prompt injection).

• Hidden Unicode characters detected (1 type(s) found)