The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's configuration templates suggest the use of the official Playwright MCP server (@playwright/mcp) for browser automation. Playwright is a well-known and trusted tool maintained by Microsoft.- [COMMAND_EXECUTION]: The skill executes browser-based commands (clicks, typing, navigation) to test web applications. These actions are governed by strict safety protocols defined in references/safety-and-scope.md, which mandate read-only behavior in production environments and forbid destructive operations such as deleting data or submitting financial transactions.- [DATA_EXFILTRATION]: Test results are transmitted to external communication platforms like Slack or Telegram as configured by the user. The reporting process includes a step to read recent channel history to prevent duplicate reports, which is a localized use of these APIs for status tracking.- [PROMPT_INJECTION]: The skill includes instructions to trigger behavior even if specific keywords aren't used, which is a standard functional instruction for agent activation and does not attempt to bypass safety filters. In fact, it reinforces safety guidelines throughout its documentation.- [INDIRECT_PROMPT_INJECTION]: Because the skill processes live website content and local repository files, it has a surface for indirect prompt injection. This risk is effectively managed through 'adversarial validation'—a process where a separate agent independently verifies findings against the source code—and explicit instructions to ignore forbidden actions regardless of external input.

monkey