The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes a workflow where the agent captures screenshots of web pages and uses a vision model (look_at tool) to analyze them for layout and visual issues. This creates a vulnerability to indirect prompt injection, where malicious instructions hidden in the visual content of a website (e.g., in text or hidden elements) could attempt to manipulate the agent's logic or subsequent actions.
Ingestion points: references/vuda-ai.md defines tools for capturing screenshots of arbitrary URLs (screenshot_url, enhanced_page_analyzer).
Capability inventory: The agent is instructed to use these screenshots as input for analysis tools like look_at and visual_comparison.
Sanitization: There is no evidence of visual filtering or content sanitization to prevent the agent from following instructions found within the processed screenshots.
[EXTERNAL_DOWNLOADS]: The skill guides users to install the Playwright framework and the VUDA MCP server from public registries. These are standard dependencies for the skill's stated purpose.
Evidence: Installation commands include pip install playwright, npm install -D @playwright/test, and npm install -g visual-ui-debug-agent-mcp.
[COMMAND_EXECUTION]: The skill documentation provides several shell commands for environment setup, browser installation, and running regression tests.
Evidence: Examples include npx playwright test, playwright install chromium, and docker run for executing tests in a consistent environment.

playwright-visual-regression