dogfood
Pass
Audited by Gen Agent Trust Hub on Jul 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
Bashtool to perform routine file system operations such as creating directories (mkdir -p) and copying templates (cp). It also executes a specialized browser automation tool,agent-browser, to interact with target web applications. - [DATA_EXFILTRATION]: While the skill captures screenshots and video recordings of the target application, these files are saved to a local output directory defined by the user. There is no evidence of these files being transmitted to unauthorized external domains.
- [PROMPT_INJECTION]: The skill is designed to navigate and interact with arbitrary external web content. This creates a surface for indirect prompt injection, where a malicious website could attempt to influence the agent's behavior through instructions embedded in HTML or text content. However, the agent's capabilities are limited to browser interactions and local file management within the context of the testing session.
- [EXTERNAL_DOWNLOADS]: The configuration metadata references
npx agent-browser, which would involve downloading packages from the public npm registry. The skill's instructions specifically advise against using this method in favor of a direct local binary for performance reasons.
Audit Metadata