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 queries the public arXiv API (http://export.arxiv.org/api/query) as documented in SKILL.md and implemented in scripts/arxiv_search.py, ingesting titles/abstracts/comments (user-submitted public content) that the agent parses, summarizes, and uses to decide follow-up actions, so untrusted third-party content could influence behavior.