docling-document-intelligence

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly accepts and converts arbitrary URLs (SKILL.md Step 1 and examples like "docling https://arxiv.org/pdf/..."), and pipelines.md plus the evaluation/chunking workflow show the agent will ingest, parse, and chunk untrusted web-hosted documents (and even send page images to remote VLM APIs), so third-party content is read and can materially influence pipeline choices, chunking/RAG inputs, and subsequent actions.