markdown-proxy

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

• Third-party content exposure detected (high risk: 0.95). Outsider-authored free text is fetched at runtime from arbitrary third-party URLs via the proxy cascade (e.g., curl -sL "https://r.jina.ai/{original_url}" / https://defuddle.md/{original_url} / npx agent-fetch "{original_url}"), and the returned page/article text is then inserted into the agent’s LLM context as “正文”.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly runs "npx agent-fetch" at runtime (Step 3), which will download and execute code from the external package (see https://github.com/teng-lin/agent-fetch), so it fetches and executes remote code during skill operation.