playwright-visual-regression
Warn
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: MEDIUMMETADATA_POISONINGEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [METADATA_POISONING]: The skill metadata lists 'mte90' as the author, which differs from the provided context attributing the skill to 'codeatcode'.
- [EXTERNAL_DOWNLOADS]: The instructions direct users to install 'visual-ui-debug-agent-mcp' from the NPM registry. This package is managed by a third-party developer and is not identified as a well-known service or vendor-owned resource. The skill also references the official Playwright drivers from Microsoft's container registry.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external URLs and passes captured visual data to AI vision models via the 'look_at' tool, creating a surface for potential injection attacks from malicious web content.
- Ingestion points: URL parameters in Playwright test scripts and VUDA tools such as 'screenshot_url' and 'enhanced_page_analyzer'.
- Boundary markers: No delimiters or explicit warnings for the model to ignore embedded instructions are included.
- Capability inventory: The skill utilizes browser automation, file writes for screenshots, vision model analysis, and shell command execution.
- Sanitization: No sanitization or validation of external web content is described.
- [COMMAND_EXECUTION]: The skill provides standard shell commands for installing dependencies and running tests, including 'pip install', 'npm install', and 'npx playwright'.
- [DYNAMIC_EXECUTION]: The skill uses 'page.evaluate()' and 'page.addStyleTag()' to inject scripts and styles into the browser context. These are used for functional tasks such as disabling animations to ensure visual stability during screenshot capture.
Audit Metadata