visa-validator
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external documents via OCR and passes the resulting text to an LLM, which presents a surface for indirect prompt injection attacks. An attacker could embed instructions in a document to influence the validation outcome.\n
- Ingestion points: User-provided folder path containing visa documents, as specified in
SKILL.md.\n - Boundary markers: Instructions do not define specific delimiters or guardrail prompts to isolate untrusted document content from agent instructions.\n
- Capability inventory: OCR extraction, file system read access for documents and checklists, and file write access for report generation.\n
- Sanitization: The skill identifies a 'PII Scrubbing Layer' for passport numbers, but lacks logic to detect or sanitize malicious natural language instructions within the documents.
Audit Metadata