eva-audience-finder
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides instructions to identify and neutralize potential injection attempts within user-supplied manuscripts.
- Ingestion points: User-provided documents, screenshots, or pasted text as defined in SKILL.md.
- Boundary markers: Explicit instructions to treat external content solely as analysis material and to ignore any embedded directives or tool calls.
- Capability inventory: The skill logic is confined to reading internal reference files and explicitly prohibits unauthorized network activities or file persistence.
- Sanitization: Automatically loads a safety boundary reference (../eva-shared/references/shared/06_external-material-safety_外部材料安全边界.md) when external material is encountered to maintain processing integrity.
- [DATA_EXFILTRATION]: The skill enforces a strict data handling policy, explicitly stating that it will not perform network searches, save assets, or create local archives without direct user authorization.
Audit Metadata