social-media-paper-triage

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

• Third-party content exposure detected (high risk: 1.00). The skill's SKILL.md workflow explicitly instructs the agent to fetch and parse content from public social media and web sources (e.g., "Step 1: Extract Content from Platform" with commands for 小红书, WeChat公众号, Twitter/X, Reddit, Bilibili, and Jina Reader), meaning untrusted, user-generated third‑party content is ingested and used to drive triage decisions and actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly executes a runtime fetch via curl to https://r.jina.ai/URL to retrieve arbitrary remote page content which is injected into the agent's context and thus can directly control prompts at runtime, making it a required external dependency for the skill.