agentation-self-driving
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates by opening and analyzing arbitrary, potentially untrusted web pages, creating a standard surface for indirect prompt injection.
- Ingestion points: Web pages are loaded using
agent-browser --headed open <url>and text/structural content is read viaagent-browser snapshot -iinSKILL.md. - Boundary markers: No boundary delimiters or system instructions are defined to tell the agent to ignore prompt injections or malicious instructions embedded in the target web page's text or HTML elements.
- Capability inventory: The skill possesses the capability to execute browser interactions (
click,fill,mouse move) and evaluate client-side JavaScript (agent-browser eval). - Sanitization: Content from webpage snapshots is parsed directly by the agent without any pre-filtering or sanitation mechanisms.
Audit Metadata