refactor-code-review

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

• Third-party content exposure detected (medium risk: 0.30). 该 skill 在“拉取已有 PR 流程”中会根据用户提供的 PR URL 执行 git fetch 以获取 PR diff（含新增/修改/删除内容）并在“PR 审视流程/逐文件分析”中对 diff/代码进行 LLM 分析，因此会读取由外部用户提交的 PR 文本内容。

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

• Potentially malicious external URL detected (high risk: 0.90). The skill instructs fetching a remote repository at runtime (e.g., https://gitcode.com/Ascend/RecSDK/pull/2303/diffs and git fetch https://gitcode.com/Ascend/RecSDK.git), which causes external repository content to be retrieved and injected into the agent's analysis context and thus can directly control the agent's prompts/behavior.