form-4797
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied disposition data, including asset descriptions and depreciation history, which are interpolated into a tax form draft deliverable. While the skill includes validation checks and structured output formatting to minimize risk, the ingestion of untrusted data from the user creates a surface for indirect prompt injection.
- Ingestion points: Prerequisites section in SKILL.md where property descriptions and historical data are requested.
- Boundary markers: The draft deliverable in SKILL.md uses markdown headers, tables, and code-block sections to delimit data.
- Capability inventory: The skill generates text drafts and provides instructions for platform-level browser automation; it does not contain local scripts, executables, or autonomous network operations.
- Sanitization: The skill includes mandatory math and sanity validation checks and explicitly instructs the agent to ask the user for confirmation rather than guessing values.
Audit Metadata