browse
Pass
Audited by Gen Agent Trust Hub on Nov 22, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads and installs the
@browserbasehq/browse-clipackage from the npm registry during the setup check if the binary is not already present. - [COMMAND_EXECUTION]: The skill uses the
Bashtool to execute various browser automation commands such asbrowse open,browse click, andbrowse typeto interact with websites. - [DYNAMIC_EXECUTION]: The
browse evalcommand allows the agent to execute arbitrary JavaScript within the context of the active browser page, which is a powerful capability used for custom data extraction and page interaction. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from external, untrusted websites, creating a surface for indirect prompt injection attacks.
- Ingestion points: Untrusted data enters the agent's context through
browse snapshot(accessibility tree),browse get text,browse get html, andbrowse screenshot(SKILL.md). - Boundary markers: There are no explicit boundary markers or instructions to the agent to ignore potentially malicious instructions embedded in the web content being processed.
- Capability inventory: The skill has access to the
Bashtool for command execution and performs network operations via the browser session (SKILL.md). - Sanitization: Content retrieved from websites is passed to the agent without sanitization or filtering of potential injection patterns.
Audit Metadata