Local-Web-Capture

Capture a list of URLs in one job. Saves each article individually with a timestamp, then produces a summary report across the batch (markdown always; PDF optional via Typst). Supports two report modes: human-summary (readable prose + per-item cards) and agent-summary (JSON optimised for LLM consumption). Use when the user supplies multiple URLs — pasted, in a file, or from clipboard — and wants both the individual captures and a synthesised overview.

Compile a batch of captured articles into a single PDF via Typst. Each article gets a header block showing the source URL and capture date, a page-numbered footer runs throughout, and a cover + table of contents open the document. Use when the user wants a printable/shareable compendium of a scrape batch, says "make a PDF of these captures", "compile the batch to PDF", or "print the batch report".

Browse, search, or summarise what's in the local-web-capture data directory (~/local-web-capture/). Use when the user asks "what did I capture recently", "find my captures of X", "list articles from ynet", or wants a table of recent captures.

Capture an article from a URL (via the normal localhost scrape path) and save a translated version alongside the original. Default language pair is Hebrew → English; supports arbitrary source → target. Saves to ~/local-web-capture/translations/. Use when the user shares a Hebrew article URL and wants the English version, says "capture and translate", or passes a URL in a language they want rendered into another.

Scrape an article from a URL via the user's localhost (preserves Israeli IP for geo-restricted sites) and save a clean markdown capture. Use when the user gives a URL and wants the article body saved, or says "capture this article", "scrape this", or references Israeli news sites (ynet, haaretz, n12, calcalist, timesofisrael, etc.).

Tier-3 escalation — capture an article or page from a paywalled, logged-in, or aggressively-gated site by driving the user's real Chrome via bb-browser (login state + cookies + user's IP). Use ONLY when scrape-article reports that headless rungs failed because of auth/paywall (e.g. Haaretz, Calcalist+, subscriber-only Israeli content). Do not use for ordinary scrapes — the headless path is cheaper.

Capture product name + price (and optional unit price, stock, SKU) from an Israeli e-commerce URL via the user's localhost, saving a structured JSON snapshot under ~/local-web-capture/prices/<domain>/. Use when the user wants to track a price on Yad2, Shufersal, Rami Levy, KSP, Ivory, Zap, or similar, or asks to "grab the price" / "snapshot this listing".

General-purpose web page scraper via the user's localhost (preserves Israeli IP for geo-restricted sites). Handles any page that is NOT a clean article — SPAs, branch locators, product catalogs, search result pages, government portals, dashboards, API endpoints discovered via devtools, or anything requiring raw HTML, rendered DOM, network/XHR responses, or structured extraction. Supports scripted user interaction (click "load more", scroll, type into search, iterate a city dropdown) for sites that hide data behind UI actions — a naive "scrape this URL" will return nothing on those. Always tries to identify and call an unauthenticated backend JSON API first — UI automation is the fallback, not the default. Canonical user invocation provides a triple: (URL, interactive target to click/type/scroll, data to extract) — the skill loads the URL, activates the target in a loop until it stops producing new content, then extracts the requested data. Trigger phrases: "scrape this page", "dump the HTML", "find the API behind this page", "capture this SPA", "get all the X from this page", "extract the JSON", "scrape branch list / store list / catalog", "load URL, click this, then extract that", "I had to click to load everything", any non-article URL. For clean article extraction use `scrape-article` instead.

First-run / re-check setup for Local-Web-Capture. Verifies the data directory exists, installs the Rung-1 headless extractor (webclaw), sets up the Rung-4 real-browser path (bb-browser), seeds sites.yaml, and flags Playwright MCP for removal once validated. Use when the user wants to initialise the plugin, when a capture skill reports a missing dependency, when onboarding a new machine, or to re-check the install.