ref-verify
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill transmits user-provided email addresses to the Unpaywall API (api.unpaywall.org) via URL query parameters. While this is a documented requirement for the service, it involves the transfer of PII to a third-party domain.
- [INDIRECT_PROMPT_INJECTION]: The skill fetches and processes research abstracts from external academic sources such as CrossRef, Semantic Scholar, and arXiv. This ingestion of untrusted content constitutes a potential attack surface for instructions embedded within research papers.
- Ingestion points: Abstracts retrieved from multiple external academic metadata providers.
- Boundary markers: The agent is instructed to use verbatim quoting, but the skill does not implement formal delimiters to isolate external text from the system prompt.
- Capability inventory: The skill is capable of network communication with academic APIs and presenting fetched data to the user context.
- Sanitization: The skill strips JATS XML tags from raw abstract data.
Audit Metadata