extract

Extract is a web-scraping skill that turns a request like "scrape the table of results" or "collect all job listings" into two concrete deliverables: a standalone, executable Playwright `.cjs` script that reproduces the extraction, and the extracted data written to disk as JSON (default), CSV, or another requested format. It addresses the fragility of naive scraping by producing scripts that account for how modern sites actually render content.

The skill uses Actionbook as a conditional accelerator: it first tries `actionbook search` and `get` to pull known-good selectors when a site is indexed with a sufficient health score, and only falls back to live browser inspection (`snapshot` for the accessibility tree, `screenshot` for visual confirmation, DOM inspection) when needed. Its core value is a mechanism-aware script strategy with worked code for the patterns that break simple scrapers: waiting for streaming/SSR/RSC hydration to complete, scroll-and-collect loops for virtualized lists, infinite-scroll with no-growth tolerance, and navigation-aware click-through pagination that only advances when the click actually caused a change. It also lays out an execution chain — identify URL, data shape, scope, and output format, then obtain selectors and choose the path.

It targets developers and data analysts who need reliable, repeatable extraction from product listings, tables, search results, feeds, or profiles. Because it drives a real browser and generates automation that scrolls, paginates, and harvests content from arbitrary sites, users should apply it within the target sites' terms of service and rate limits.