research

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

• Third-party content exposure detected (high risk: 1.00). The SKILL.md Subagent prompt contract and tool chain explicitly require using TinyFish fetch and Firecrawl scrape/crawl to retrieve and read arbitrary public URLs (web pages, blogs, papers) and then use those fetched contents as evidence to drive synthesis and follow-up actions, which exposes the agent to untrusted third‑party content that could inject instructions.

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

• Potentially malicious external URL detected (high risk: 0.70). The skill's scripts call https://alphaxiv.org (scripts/alphaxiv.sh uses curl to fetch /overview/{id}.md and /abs/{id}.md at runtime) and those AI-structured summaries are injected into the agent's evidence/context for papers, meaning external content from https://alphaxiv.org can directly shape agent prompts/outputs.