cross-pollinator

Pass

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADS
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches scientific data from the official arXiv API (export.arxiv.org). This is a well-known service and the usage is consistent with the skill's research-oriented purpose.
  • [PROMPT_INJECTION]: The skill demonstrates an attack surface for indirect prompt injection as it processes untrusted external content from paper abstracts. However, the risk is negligible due to the following factors:
  • Ingestion points: Data enters the context via the arXiv API in Step 2 of the methodology.
  • Boundary markers: The instructions do not define specific delimiters or instructions to ignore embedded commands within the fetched abstracts.
  • Capability inventory: The skill does not have any dangerous capabilities, such as file system writes, subprocess execution, or network exfiltration to arbitrary domains.
  • Sanitization: No explicit sanitization or filtering of the external data is described.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 16, 2026, 02:32 AM
Security Audit — agent-trust-hub — cross-pollinator