agents

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

• Third-party content exposure detected (medium risk: 0.30). complexity-triage.md 在運行時會讀取「需求文本（含 Notion 票、對話脈絡）」作為資料進行分診，且存在外部使用者可提交/發佈需求文本的路徑而在未先選定特定條目前就會被模型讀入。

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly falls back at runtime to fetching agent definitions from https://github.com/ashe-li/agent-skills when agents/.md is missing, which would load remote content that directly controls agent prompts.