arxiv-search

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

• Third-party content exposure detected (high risk: 0.70). The skill explicitly fetches and parses public arXiv content via the export.arxiv.org API (see _search_arxiv_once and _best_effort_enrich in scripts/run.py and the SKILL.md "online" retrieval flow), so untrusted third‑party text (titles/abstracts/comments) is ingested and can influence filtering and downstream decisions.