web-content-fetcher

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

• Third-party content exposure detected (high risk: 1.00). SKILL.md and scripts/fetch.py explicitly fetch and extract content from arbitrary public URLs (e.g., mp.weixin.qq.com, blogs, news sites and via Jina Reader's r.jina.ai) so the agent consumes untrusted, user-generated third‑party web content as part of its workflow, which could carry indirect prompt injections.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill runs at runtime to fetch arbitrary external pages via python3 scripts/fetch.py "" (and may call the Jina Reader fallback at https://r.jina.ai/), and the retrieved content is injected/returned for the agent to use—meaning remote content can directly influence prompts (prompt‑injection) and thus control agent behavior.