research-ideation

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly performs WebSearch of open web sources (arXiv, Google Scholar, landing pages) and then imports and reads those landing pages/PDFs via the Zotero MCP tools (e.g., zotero_add_items_by_identifier, PDF cascade, zotero_get_item_fulltext), so it ingests untrusted public third‑party content that the agent analyzes and uses to drive research decisions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill's Zotero integration explicitly fetches external papers/PDFs at runtime (e.g., via https://arxiv.org/abs/...), and these full-texts are read with zotero_get_item_fulltext and incorporated into the agent's analysis/context, meaning remote content can be injected into and directly influence prompts.