claude-browser

Pass

Audited by Gen Agent Trust Hub on Apr 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection vulnerability surface.
  • Ingestion points: Data enters the agent context through read_page, get_page_text, read_console_messages, and read_network_requests tools in SKILL.md.
  • Boundary markers: The skill does not define delimiters or provide instructions to ignore commands embedded in the ingested web content.
  • Capability inventory: The skill has access to powerful tools including javascript_tool (arbitrary JS execution), form_input (user interaction), computer (mouse/keyboard control), and upload_image (data capture).
  • Sanitization: No sanitization or validation of the ingested external content is described.
  • [COMMAND_EXECUTION]: The javascript_tool allows the agent to execute arbitrary code within the context of the user's browser tabs, providing a vector for exploitation if the agent is misled by malicious data.
  • [DATA_EXFILTRATION]: The skill targets authenticated web applications and provides tools (read_page, read_network_requests, upload_image) that could be used to extract sensitive session data, cookies, or personal information if manipulated.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 13, 2026, 11:49 AM