Playwright (Automation + MCP + Scraper)

**Judgement:** This skill is a strong, opinionated layer around Playwright and Playwright MCP, well-suited for serious browser automation, UI testing, and debugging when a real browser is actually needed. It prioritizes robust selectors, isolation, and CI-friendly patterns over quick-and-dirty scripts. **Recommended scenarios:** - Automating JS-heavy or authenticated flows where static HTTP fetch is not enough. - Repo-owned browser work: Playwright test suites, regression repros, CI debugging, screenshots/PDFs/traces as artifacts. - Agent workflows that already use `browser_*` MCP tools and need reliable navigate–click–fill–screenshot flows. - Occasional rendered-page scraping where you explicitly accept browser cost/complexity. **Key strengths:** - Clear guidance on when *not* to use a browser (prefer HTTP-first), reducing cost and brittleness. - Strong best practices: resilient locators, actionability-based waits, isolation of runs, and explicit handling of auth/state. - Good coverage of debugging and CI (traces, logs, environment drift) and integration with existing Playwright configs. **Risks & limitations:** - Any Playwright/MCP-driven browser automation inherits typical fragility: dynamic DOM changes, third-party widgets, and flaky upstream services can still cause failures. - Performance and resource usage are higher than HTTP-only scraping, so this is ill-suited as a default scraper for large-scale crawling. - Safety around production and high-stakes flows depends on user discipline; while the skill advises confirmation and staging use, it can still drive real production sessions if asked. **Bottom line:** Use this skill when you **truly need a real browser** for automation, tests, or debugging and want structured, best-practice Playwright/MCP usage rather than ad-hoc scripts. Prefer simpler HTTP-based skills for basic scraping or data retrieval where a full browser is unnecessary.