external-gitcode-ascend-atb-aclnn-operator-replacement-designer

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

• Third-party content exposure detected (high risk: 0.85). 该技能在“接口文档分析”阶段会对用户提供的 ATB/ACLNN 接口 URL 执行 WebFetch，并将抓取到的网页/文档正文作为可读文本用于 LLM 上下文；这属于“公共 web 内容/运行时抓取的外部自由文本”路径。

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

• Potentially malicious external URL detected (high risk: 0.90). The skill requires fetching two user-provided interface-document URLs at runtime via WebFetch (the "ATB 算子接口文档链接" and "ACLNN 算子接口文档链接") and mandates that both fetches succeed because the fetched document contents are parsed and directly drive the generated prompts/instructions for the design output (user-provided ATB and ACLNN interface document URLs fetched via WebFetch).

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

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