paper-orchestra

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

• Third-party content exposure detected (high risk: 1.00). The skill's Literature Review step explicitly performs web searches and Semantic Scholar fetches (see SKILL.md, references/host-integration.md and pipeline.md) to ingest public third‑party papers into citation_pool/intro_relwork.tex and refs.bib which are then read and used by the Section Writing and downstream agents, so external untrusted content can directly influence tool decisions and generated outputs.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly calls Semantic Scholar at runtime (e.g., https://api.semanticscholar.org/graph/v1/paper/...) for citation verification and uses the fetched metadata/abstracts to build the verified citation pool that is then injected into LLM prompts to draft the Intro/Related Work, so this external URL directly controls prompt content during execution.