paper-card-delivery
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates a workflow for ingesting and summarizing external, untrusted academic research papers, creating a potential surface for indirect prompt injection.
- Ingestion points: External research paper sources including official PDFs, arXiv HTML, publisher/conference pages, and project websites as defined in the 'Required Source Order' section of
SKILL.md. - Boundary markers: The instructions emphasize a 'Core Contract' for source verification and provide a detailed 'Verification Checklist' to ensure all content is grounded in the source. While these guide the agent's behavior, there are no technical delimiters (like XML tags or specific 'ignore' markers) defined for the data itself during processing.
- Capability inventory: The skill writes content to Feishu/Lark, Notion, and Obsidian. It utilizes local Python scripts for validation and references internal tools like
mineru-md.shfor PDF extraction. - Sanitization: There is no evidence of programmatic sanitization or payload filtering for content extracted from external papers; the skill relies on the agent's reasoning-based verification and structural validation scripts.
Audit Metadata