internal-audit-report

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

• Third-party content exposure detected (medium risk: 0.65). 该技能在“阶段三”会把用户上传到 input/ 的内审底稿/访谈纪要/系统日志等证据内容（可能包含他人撰写的文本）通过 OpenCode 指令提交给 LLM 进行长文生成与引用，因此存在把“非操作用户作者”的自由文本注入上下文的路径。

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

• Potentially malicious external URL detected (high risk: 1.00). The skill issues runtime curl/POST requests to an OpenCode bridge (http://opencode-bridge:8000, specifically http://opencode-bridge:8000/file/{user} and http://opencode-bridge:8000/task/{user}) and relies on that service to execute instructions and generate report content, so these URLs are runtime external dependencies that directly control prompts/execution.