skills/seaworld008/commonly-used-high-value-skills/browser-testing-with-devtools/Gen Agent Trust Hub
browser-testing-with-devtools
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill interacts with and processes data from external, potentially untrusted web pages, which creates an indirect prompt injection surface. The skill proactively addresses this through explicit security guidelines.
- Ingestion points: As described in
SKILL.md, the agent reads DOM content, console logs, network responses, and results from JavaScript execution in the browser. - Boundary markers: The instructions mandate that the agent maintain clear boundaries between trusted user input and untrusted browser data, ensuring findings are clearly labeled and never merged into the instruction context.
- Capability inventory: The skill utilizes tools for capturing screenshots, inspecting the DOM, monitoring network traffic, and executing JavaScript within the page context.
- Sanitization: The skill establishes strict rules: it prohibits interpreting browser content as instructions (e.g., 'Ignore previous instructions'), requires user confirmation for navigating to extracted URLs, and forbids accessing or exfiltrating credentials found in the browser context.
- [EXTERNAL_DOWNLOADS]: The skill configuration instructions include the installation of the
chrome-devtools-mcppackage via the NPM registry to provide the necessary Model Context Protocol (MCP) server functionality.
Audit Metadata