Find an agent skill by task, domain, or provider.

Use when an agent needs to run browser regression or verification checks: run them as Playwright specs through the CLI (`playwright test -c <config> --reporter=line`), write JSON results outside the checkout, read only the failures, and open a trace only when a test fails. Do not drive multi-step checks through the built-in browser pane, Claude in Chrome or a Playwright MCP; use an interactive browser only to explore or find selectors. Run Playwright configs that boot their own `next dev` one at a time unless each has its own checkout clone and port. Materialized from a reviewed skills.sh missing-skill backfill. Candidate provenance is stored in metadata.