playwright-testing

Pass

Audited by Gen Agent Trust Hub on Mar 1, 2026

Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill leverages Playwright MCP tools to perform browser interactions like clicking, typing, and navigating. This is standard functionality for automated testing skills.\n- [REMOTE_CODE_EXECUTION]: Through the use of mcp__playwright__browser_evaluate, the skill executes dynamic JavaScript within the target browser environment. While necessary for deep testing of canvas/WebGL applications, it represents a form of remote code execution on the target site.\n- [EXTERNAL_DOWNLOADS]: The imgdiff.py script included in the skill references the pillow library. Pillow is a well-known, trusted, and widely-used Python package for image processing.\n- [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests and reasons over data retrieved from external, potentially untrusted web environments.\n
  • Ingestion points: Untrusted data is ingested via mcp__playwright__browser_console_messages, mcp__playwright__browser_snapshot, and state data returned by mcp__playwright__browser_evaluate.\n
  • Boundary markers: The instructions do not specify any boundary markers or delimiters to help the agent distinguish between its own instructions and the data extracted from the browser.\n
  • Capability inventory: The agent has the capability to navigate to arbitrary URLs, click/type on pages, and execute arbitrary JavaScript in the browser.\n
  • Sanitization: There is no explicit sanitization or validation of the data extracted from the browser before it is passed to the agent for analysis.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 1, 2026, 03:30 PM