paper-digest

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

• Third-party content exposure detected (low risk: 0.10). In SKILL.md, the required runtime workflow fetches outsider-authored free text from arXiv by first retrieving paper_text via web.fetch https://arxiv.org/html/<arxiv_id> (and later fetching citation pages via web.fetch https://arxiv.org/html/<citation_id>), where the fetched HTML content is authored by third parties and not pre-selected from a trusted first-party source list.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill fetches remote arXiv HTML at runtime (e.g., https://arxiv.org/html/<arxiv_id> and variants) and injects that fetched content into the agent's workspace/context to drive summarization, so external content directly influences the agent's prompts/output.