/scrape — Read-only web data extraction

One-line intent in, one JSON document out.

Step 1 — Determine intent

The text after /scrape is the intent. If absent, ask once: "What do you want to scrape? One line, e.g. 'top stories on Hacker News' or 'product names + prices on example.com/products'." Don't front-load clarifying questions — cheaper to refine while prototyping.

Step 2 — Refuse mutating intents

If the intent implies writes — submit, post, send, log in, click X, fill the form, delete, create, order, book — refuse:

"/scrape is read-only. For mutating flows, drive the page directly with Playwright or ask for a dedicated automation script." Stop. Do not enter the prototype path.

Step 3 — Check for an existing saved scraper

Look in scrapers/ (repo root) for a saved scraper whose intent/triggers match (written by /skillify). If scrapers/<name>.mjs exists and clearly covers this intent:

node scrapers/<name>.mjs   # prints the JSON directly

Emit its JSON and stop. If no confident match, fall through to the prototype path.

Step 4 — Prototype path

Pick the lightest tool that works:

A. Static HTML (default — try first): use the WebFetch tool on the URL with a prompt that asks for the specific data as JSON. Good for server-rendered pages.

B. JS-rendered / WebFetch came back empty or partial: use Playwright via npx. Write a throwaway probe, iterate selectors against the real HTML:

cat > /tmp/scrape-$$.mjs <<'EOF'
import { chromium } from 'playwright';
const url = process.argv[2];
const b = await chromium.launch();
const p = await b.newPage();
await p.goto(url, { waitUntil: 'networkidle', timeout: 30000 });
const html = await p.content();
await b.close();
process.stdout.write(html);
EOF
npx -y playwright@latest /tmp/scrape-$$.mjs "<url>" > /tmp/scrape-$$.html

Then read /tmp/scrape-$$.html, find the repeating structure (list rows, table cells, cards), and extract. For lists of links, pull a[href]. Iterate: try a selector, check the output, refine. (Run npx -y playwright@latest install chromium once if the browser is missing.)

Step 5 — Emit

One JSON document on stdout. Use a stable shape — typically { "items": [...], "count": N }. Do not wrap it in prose in the chat reply (callers pipe to jq) unless the user asked for an explanation. Logs and notes go in chat, not in the JSON.

Step 6 — Skillify nudge

After a successful prototype, append exactly one line and nothing more:

"Say /skillify to save this as a reusable scraper (instant on next call)."

When the prototype fails

If after 3–4 selector attempts no sensible JSON shape emerges:

• Report what you tried, what came back, what's blocking (lazy-loaded, JS-gated, paywalled, bot-blocked).
• Do NOT write a partial result and call it done. Do NOT suggest /skillify on a broken prototype.
• Ask: (a) try a different selector, (b) different page, or (c) stop.

What this skill does NOT do

• Mutating actions, auth/login flows, cookie import, multi-page crawls. One-shot, read-only, per call.

Respect & limits

• Honor robots.txt intent and rate limits; this is for data the user is entitled to read.
• Never scrape anything behind a login the user hasn't authorized you to use.