argent-test-ui-flow
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and acts upon text data extracted from external applications under test, introducing a potential indirect prompt injection surface if the app under test renders untrusted or malicious content.
- Ingestion points: Application screens, components, and text attributes are ingested via tools like
describe,debugger-component-tree, andnative-describe-screenwithin the testing workflow inSKILL.md. - Boundary markers: Absent. There are no guidelines or explicit delimiters instructing the agent to isolate or treat text found inside the app's UI as untrusted content.
- Capability inventory: The skill possesses interactive capabilities including execution of
gesture-tap,gesture-swipe,keyboard, and platform buttons that are dynamically driven by the results of UI discovery. - Sanitization: Absent. No sanitization, escaping, or strict validation of structural text elements is described before the agent interacts with the UI based on those inputs.
Audit Metadata