copywriting

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

• Third-party content exposure detected (medium risk: 0.65). 该技能在运行时会把用户提供的 materialText/materialFile/materialUrl（尤其是 materialUrl 指向的公开网页内容）作为可读文本进入“contextComposer”并送入 LLM；若这些内容来自非操作用户（外部网页/他人文案），则属于 OUTSIDER 免费文本经由 materialUrl/materialText 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: 1.00). The provided runner makes runtime POST requests to the external execution endpoint https://ai-skills.ai (via DEFAULT_BASE_URL or AISKILLS_BASE_URL) to run the skill, so this remote URL is required at runtime and directly controls/executes the skill logic and prompts.