prompts.js

// LICENSE_CODE ZON
'use strict'; /*jslint node:true es9:true*/

const web_scraping_strategy = {
    name: 'web_scraping_strategy',
    description: 'Decision tree for picking the right Bright Data tool. '
        +'Invoke at the start of any scraping session to learn the correct '
        +'tool selection order '
        +'(pre-built scrapers -> Web Unlocker -> Browser API).',
    arguments: [],
    load: ()=>`\
You have access to Bright Data tools at three tiers of cost and capability.
Always follow this order -- do not skip ahead:

STEP 1 -- Check for a pre-built scraper (fastest, structured JSON):
  Call search_scrapers with the site or data type (e.g. "amazon product",
  "linkedin.com"). If one matches, call get_scraper_details, then
  run_scraper with input shaped like sample_input.

STEP 2 -- If no scraper matches, use scrape_as_markdown (default):
  This handles anti-bot protection and CAPTCHA automatically.
  Retry once if the first attempt returns empty or blocked content.

STEP 3 -- If scrape_as_markdown fails twice, escalate to
scraping_browser_navigate:
  Use ONLY when the page requires JavaScript execution, user interaction
  (clicking, form submission), or dynamic content loading.
  This is slower -- do not use as a first attempt.

NEVER use the browser tools for sites scrape_as_markdown can handle.
NEVER use scrape_as_markdown when a pre-built scraper covers the site.`,
};

const diagnose_scraping_approach = {
    name: 'diagnose_scraping_approach',
    description: 'Run a two-step diagnostic to discover the correct '
        +'Bright Data product for a new website. Tries Web Unlocker first, '
        +'then Browser API, then reports which succeeded.',
    arguments: [],
    load: ()=>`\
To discover the correct Bright Data product for a new website, run this
diagnostic:

1. Try scrape_as_markdown on the target URL.
   - If it returns useful content -> Web Unlocker is the correct
     integration. Stop.
   - If it returns empty, blocked, or low-quality content -> continue
     to step 2.

2. Try scraping_browser_navigate + scraping_browser_snapshot on the
   same URL.
   - If it returns useful content -> Browser API is the correct
     integration. Stop.
   - If both fail -> report to the user that the target may require a
     specialized Bright Data product (SERP API, a pre-built scraper,
     or custom configuration).

Report which approach succeeded and recommend it as the integration method.
Do not proceed with data extraction until the diagnostic is complete.`,
};

const prompts = [web_scraping_strategy, diagnose_scraping_approach];

export default prompts;