claude-in-chrome
Warn
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: MEDIUMINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to automate web browsing and read content from external web pages, creating a surface where malicious instructions embedded in web content could be processed by the agent.
- Ingestion points:
mcp__claude-in-chrome__read_page,mcp__claude-in-chrome__read_console_messages, andmcp__claude-in-chrome__read_network_requests(SKILL.md). - Boundary markers: No explicit instructions for delimiters or safety warnings regarding external content are present in the skill instructions.
- Capability inventory:
mcp__claude-in-chrome__javascript_tool(JS execution),mcp__claude-in-chrome__navigate,mcp__claude-in-chrome__tabs_create_mcp, andmcp__claude-in-chrome__form_input(SKILL.md). - Sanitization: No evidence of sanitization or filtering of data read from the browser is provided in the instructions.
- [DYNAMIC_EXECUTION]: The skill explicitly instructs the agent to use the
mcp__claude-in-chrome__javascript_toolfor 'page scripting' and for interacting with page elements. This allows for the generation and execution of arbitrary JavaScript code within the context of the user's browser session at runtime.
Audit Metadata