The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes bundled Node.js scripts (visual_layout_audit.mjs and silent_degradation_probe.mjs) to automate browser interactions and analyze UI rendering. This is the core functionality of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill audits external URLs and local HTML files provided by the user, which constitutes an ingestion of untrusted data.
Ingestion points: Content is loaded via Playwright through the --url and --file arguments in the provided scripts.
Boundary markers: The instructions do not define specific text boundaries, but the agent's task is limited to visual and structural auditing.
Capability inventory: The skill has the ability to launch a browser (Playwright), read local files, and write report/screenshot data to the filesystem.
Sanitization: Both scripts include robust redaction logic that strips sensitive information—such as path secrets, query parameters, and fragments—from URLs before they are included in reports or logs. Additionally, text content in reports is hashed to avoid logging PII/secrets.
[DYNAMIC_EXECUTION]: The audit scripts dynamically resolve and import the playwright library from the audited project's environment. This is a standard and documented approach to leverage the existing test infrastructure of the project being audited. The scripts also utilize page.evaluate() to execute diagnostic JavaScript within the browser context of the target page, which is a fundamental part of the auditing process.

frontend-visual-qa