paper-graph

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.70). Most URLs are official academic APIs (arXiv, Semantic Scholar, merlot/mermaid, doi.org) and appear benign, but the list includes an obscure third‑party arXiv fallback domain (markxiv.org) that resembles arxiv.org (possible typosquat/unvetted content) and a GitHub repo under a cryptic username (qhjqhj00) — both are unvetted third‑party sources that raise moderate suspicion for content-distribution risk.

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

• Third-party content exposure detected (medium risk: 0.30). Outsider-authored free text (the user’s research query, written at Step 1 and then read by Step 4/8 prompt assembly) is ingested by the host LLM to generate search phrases, which are then used by the runtime workflow’s deterministic fetchers to pull paper abstracts/sections from Semantic Scholar/DeepXiv and include them in later LLM steps.