scibox-diagram

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

• Third-party content exposure detected (medium risk: 0.30). Skill.md 的运行流程在本地渲染脚本（如 scripts/roadmap_5band.py / framework_3col.py / stageflow_3col.py / taskflow_land.py）中直接读取用户提供的 content.json 文本并写入到 .drawio XML 的 mxCell value 字段（通过 f'...{need(d,...)}...' 等路径），因此外部可控文本会被 LLM/运行时消费参与生成。

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

• Potentially malicious external URL detected (high risk: 0.80). The preview_html script embeds and opens https://embed.diagrams.net/?embed=1&proto=json&spin=1&ui=atlas&libraries=0&grid=0 at runtime (and checks origin https://embed.diagrams.net), which causes remote JavaScript from that domain to be fetched and executed in the user's browser during the skill's preview operation.