job-fin
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external websites and shared documents, creating a potential surface for indirect prompt injection.\n
- Ingestion points: Financial data is ingested from the Juchao Information website (Scenario 3) and internal business spreadsheets or approval flows (Scenarios 1, 2, and 4).\n
- Boundary markers: The instructions explicitly direct the agent not to invent missing information, to flag insufficient data rather than using defaults, and to confirm calculation formulas before execution.\n
- Capability inventory: The skill reads web content, internal spreadsheets, and platform-specific approval records.\n
- Sanitization: The skill mandates that all outputs must be manually reviewed and cross-referenced with original source documents before use.\n- [EXTERNAL_DOWNLOADS]: The skill retrieves financial reporting data from Juchao Information (cninfo.com.cn), which is a well-known financial disclosure service.\n
- Evidence: Scenario 3 instructs the agent to extract annual report data from '巨潮资讯网' (Juchao Information).
Audit Metadata