The Agent Skills Directory

[COMMAND_EXECUTION]: Utilizes npx playwright test as a fallback mechanism for performing browser-based UI verification and interaction testing when MCP tools are unavailable.
[COMMAND_EXECUTION]: Invokes the open command on macOS to display the generated HTML review reports to the user immediately after generation.
[PROMPT_INJECTION]: The skill ingests untrusted data from user-provided URLs via browser navigation and accessibility snapshots. This creates a surface for indirect prompt injection where a malicious website could attempt to influence the agent's analysis or output.
Ingestion points: External URLs accessed via browser_navigate as described in playwright.md and SKILL.md.
Boundary markers: No explicit boundary markers or instructions to ignore embedded content are defined for the data retrieved from pages.
Capability inventory: The skill uses npx playwright for browser control and interaction (click, type, etc.) and open for file access, as documented in SKILL.md and playwright.md.
Sanitization: There is no mention of sanitizing or escaping the data extracted from the web pages (e.g., accessibility trees or console logs) before it is processed by the agent.

ui-ux