systematic-literature-review

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly queries the public arXiv API (via scripts/arxiv_search.py as required in Phase 2 of SKILL.md) and feeds untrusted, user-submitted paper abstracts into subagents (Phase 3) whose outputs directly drive synthesis and tool actions, so arbitrary third‑party content from arXiv can materially influence agent behavior.