form-2290
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by design, as it ingests untrusted user-provided information and interpolates it into automated tool workflows for tax filing.
- Ingestion points: Legal business name, EIN, business address, and vehicle VINs are collected from the user as prerequisites in
SKILL.md. - Boundary markers: The instructions do not specify explicit delimiters or "ignore embedded instructions" warnings for the user-supplied data before it is processed by the browser automation tool.
- Capability inventory: The skill utilizes browser automation (
agent-browser) to interact with third-party e-file provider portals as described infiling.md. - Sanitization: The skill implements validation logic, including 17-character VIN checks and EIN propagation delay verification, which reduces the risks associated with processing external content.
Audit Metadata