link-drop-pipeline

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

• Third-party content exposure detected (high risk: 0.85). SKILL.md Step 1/2/4 describes fetching outsider-authored web content (tweets, GitHub READMEs, blog text, PDFs, etc.) and then using that extracted text as input to downstream LLM-driven extraction/routing, which creates an indirect prompt-injection path into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill fetches arbitrary external content at runtime (e.g., X/tweet URLs such as https://x.com/rubenhassid/status/2045004326502871429 and https://x.com/akshay_pachaar/status/2045510648474530263), and that fetched content is used to generate new skills/agents (i.e., directly controls the agent's produced instructions/artifacts), so these runtime URL fetches present a high-risk vector.