web-debug-browser
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill documentation (references/operations.md, references/troubleshooting.md) recommends using
npx chrome-devtools-mcp@latest, which fetches and executes the vendor's MCP server package from the NPM registry at runtime. - [DYNAMIC_EXECUTION]: The
evaluate_scripttool enables the execution of arbitrary JavaScript functions within the debugged browser page (references/tool-reference.md). This is a primary feature for web automation but constitutes a dynamic code execution vector. - [INDIRECT_PROMPT_INJECTION]: The skill interacts with arbitrary external web pages, creating a vulnerability surface for indirect prompt injection.
- Ingestion points: Untrusted data is retrieved from web pages via
take_snapshot,evaluate_script,list_console_messages, andlist_network_requests(SKILL.md, references/operations.md). - Boundary markers: The instructions do not provide explicit delimiters or warnings to prevent the agent from following instructions embedded within the processed web content.
- Capability inventory: The agent possesses high-privilege tools, including arbitrary script execution (
evaluate_script), local file reading for upload (upload_file), and writing files to the local disk using thefilePathparameter intake_screenshot,take_snapshot, and performance tracing tools (references/tool-reference.md). - Sanitization: No validation or sanitization of data returned from the browser context is documented.
- [DATA_EXFILTRATION]: The
list_network_requests,get_network_request, andlist_console_messagestools allow the agent to inspect network traffic and developer logs, which may contain sensitive information such as authentication headers, session cookies, or secrets logged by the application (references/tool-reference.md). - [COMMAND_EXECUTION]: The troubleshooting documentation (references/troubleshooting.md) suggests using an SSH command (
ssh -N -L ...) to establish port forwarding, which involves executing shell commands to tunnel traffic for remote debugging.
Audit Metadata