The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The local execution entry point (scripts/run-tests.sh) utilizes npx playwright test, which may download the Playwright framework and browser binaries from official repositories at runtime if not already present. This is expected behavior for a Playwright-focused automation skill.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process project context and user requirements. While this technically creates a surface for indirect prompt injection via untrusted input data, the skill instructions are focused on design and planning tasks, and there are no signs of unsafe interpolation into high-privilege execution contexts.\n- [SAFE]: The fixture server (evals/fixtures/checkout-app/server.mjs) used for testing ensures security by using a hardcoded lookup table for file serving, which prevents directory traversal or arbitrary file access attacks within the local environment.\n- [SAFE]: The evaluation verification logic (evals/fixtures/scripts/check_playwright_output.sh) uses node --check to validate the syntax of model-generated code. This approach ensures the output is valid JavaScript without executing potentially malicious or untrusted logic.

ui-test-playwright