Agentic Web Scraper Expert (2026 Edition)

English

Orchestration & Integration

Upstream Orchestrators: Works under brainstorming and zero-to-prod-orchestrator (Phase 1 & Phase 5 - Extraction & Data Pipelines) to gather web data.
Downstream Cooperating Skills:
- browser-automation-expert — Hardcore anti-bot evasion, stealth Playwright execution, and Cloudflare bypass.
- ai-llm-integration-expert — LLM schema-driven structured data extraction from DOM markdown.
- deep-research-analyst — Feed extracted web sources to multi-source evidence synthesis.
- data-pipeline-etl-expert — Ingest extracted datasets into transformation and analytics storage.
- session-memory-manager — Cache scraped web structures and extraction progress checkpoints.

Description

Advanced Agentic Web Scraping utilizing modern multi-strategy data extraction. Leverages Crawl4AI v4 and Firecrawl to convert raw DOMs into LLM-friendly Markdown. Implements Agentic Extraction loops where the LLM guides the scraper dynamically based on page state. Incorporates strategies for bypassing anti-bot measures (Cloudflare Turnstile, Datadome) and navigating dynamic Shadow DOMs.

Trigger Conditions

Extracting structured data from websites for analysis, training data, or content pipelines.
Scraping dynamic JavaScript-rendered pages and complex SPAs.
Converting web pages to clean Markdown for LLM context or RAG pipelines.
Dealing with anti-bot protections or complex Shadow DOM architectures during scraping.
Implementing an automated agentic data extraction loop.

Extracting DOM into LLM-Friendly Markdown

Use Crawl4AI v4 for high-performance async extraction and Firecrawl for seamless LLM-ready conversion.

Crawl4AI v4 (Async Python):

import asyncio
from crawl4ai import AsyncWebCrawler, BrowserConfig, CrawlerRunConfig, CacheMode

async def extract_markdown(url: str):
    config = BrowserConfig(headless=True, bypass_csp=True)
    run_config = CrawlerRunConfig(
        cache_mode=CacheMode.ENABLED,
        remove_overlay_elements=True,
        word_count_threshold=50
    )
    
    async with AsyncWebCrawler(config=config) as crawler:
        result = await crawler.arun(url=url, config=run_config)
        # Returns clean, AI-optimized markdown ready for LLM consumption
        return result.markdown.fit_markdown

Firecrawl (Managed API):

from firecrawl import FirecrawlApp
from pydantic import BaseModel

app = FirecrawlApp(api_key="fc-xxxx")

class ExtractionSchema(BaseModel):
    title: str
    content: str
    key_metrics: list[str]

# Single API call to extract structured data based on JSON schema
result = app.scrape_url(
    "https://example.com/data",
    formats=["extract", "markdown"],
    extract={"schema": ExtractionSchema.model_json_schema()}
)
print(result.markdown) # Clean markdown
print(result.extract)  # Structured JSON

Anti-Bot Bypass & Shadow DOMs

Scraping modern web apps requires bypassing anti-bot measures like Cloudflare Turnstile and Datadome, as well as accessing deeply nested elements.

Anti-Bot Bypass (Cloudflare Turnstile, Datadome):
- Residential Proxies: Rotate high-quality residential IPs to avoid datacenter IP bans.
- Browser Fingerprinting: Use tools like playwright-stealth or specialized stealth browsers (e.g., Undetected ChromeDriver, Curl-Impersonate) to mask automated fingerprints (WebGL, Canvas, User-Agent).
- Human-like Interaction: Introduce random delays, simulate realistic mouse movements, and handle CAPTCHAs via third-party solving services only when necessary.
Dynamic Shadow DOMs:
- Use CSS piercing selectors or JavaScript execution to penetrate the Shadow Root.
- Example (Playwright): await page.locator('my-web-component >> css=.internal-element').text_content()
- Recursively traverse the DOM tree injecting scripts to extract content from encapsulated components.

Agentic Extraction Loops

Implement an autonomous loop where an LLM guides the scraper based on the current page state, rather than relying on brittle CSS selectors.

Observe: The scraper extracts the current DOM into clean Markdown.
Analyze: The LLM analyzes the Markdown to identify necessary data or the next interaction step (e.g., "Click the 'Load More' button").
Act: The LLM issues a command (extract data, navigate, click, fill form).
Loop: Repeat until the extraction goal is met.

async def agentic_scrape_loop(url: str, goal: str):
    current_url = url
    while True:
        markdown_content = await extract_markdown(current_url)
        # LLM analyzes state and decides next action
        action = await llm_decide_action(markdown_content, goal)
        
        if action.type == "COMPLETE":
            return action.extracted_data
        elif action.type == "CLICK":
            await click_element(action.target_selector)
        elif action.type == "NAVIGATE":
            current_url = action.new_url

Ethical Scraping Checklist

Check robots.txt and respect Disallow rules.
Implement rate limiting.
Use descriptive User-Agent headers.
Do not scrape personal/private data without consent.

Sovereign Execution, Token Efficiency & Invariants

Dimension	Sovereign Mandate & Deterministic Contract
Frontier Model Cascade	Medium / High (`GPT-5.6 Sol Medium`, `Gemini 4 Argon`, `Claude Sonnet 5.5 High`, `Grok 4.7 High Fast`)
Availability Fallback	`Gemini 4 Argon/Pro` -> `Gemini 3.8 Flash`; `Claude Opus 5.5` -> `Claude Opus 4.6` (Zero unhandled missing-model errors)
KV-Cache & Token Protocol	Pin static instructions & schemas at prompt prefix (`kv-cache-prefix-optimizer`); pass surgical `CONTEXT_MAP.md` slices (<250 tokens/call) during Swarm Fan-Out
SaaT CLI & MCP Dispatch	CLI: `vibes skill exec web-scraper <audit\|eval\|verify\|generate\|blueprint>` / MCP: `vibes_skill_dispatch`, `vibes_ast_query`, `vibes_compile_cache`, `vibes_optimize_budget`, `vibes_verify_gate`
NFR & Anti-Slop Invariants	P99 API Gateway Latency < 85ms, OpenAPI 3.1 / TypeSpec strict schema parity, RFC 9457 Problem Details error contracts, Idempotency-Key enforcement; 100% complete zero-placeholder output verified via `node scripts/check-anti-slop.js --strict`

Bahasa Indonesia

Integrasi Orkestrasi

Orkestrator Hulu: Beroperasi di bawah arahan brainstorming dan zero-to-prod-orchestrator (Fase 1 & Fase 5 - Ekstraksi & Pipeline Data) untuk mengumpulkan data web.
Skill Hilir & Kerja Sama:
- browser-automation-expert — Bypass proteksi anti-bot, eksekusi stealth Playwright, dan Cloudflare bypass.
- ai-llm-integration-expert — Ekstraksi data terstruktur berbasis skema LLM dari DOM markdown.
- deep-research-analyst — Menyalurkan sumber web terekstraksi ke sintesis bukti multi-sumber.
- data-pipeline-etl-expert — Memasukkan dataset ekstraksi ke pipeline transformasi dan analitik.
- session-memory-manager — Menyimpan cache struktur web dan checkpoint progres ekstraksi.

Deskripsi

Scraping Web Agentic tingkat lanjut menggunakan ekstraksi data multi-strategi modern. Memanfaatkan Crawl4AI v4 dan Firecrawl untuk mengubah DOM mentah menjadi Markdown yang ramah LLM. Mengimplementasikan loop Ekstraksi Agentic di mana LLM memandu scraper secara dinamis berdasarkan status halaman. Menggabungkan strategi untuk melewati tindakan anti-bot (Cloudflare Turnstile, Datadome) dan menavigasi Shadow DOM yang dinamis.

Kondisi Pemicu

Mengekstrak data terstruktur dari situs web untuk analisis, data pelatihan, atau pipeline konten.
Scraping halaman yang dirender JavaScript secara dinamis dan SPA kompleks.
Mengonversi halaman web menjadi Markdown bersih untuk konteks LLM atau pipeline RAG.
Menghadapi perlindungan anti-bot atau arsitektur Shadow DOM yang kompleks saat scraping.
Mengimplementasikan loop ekstraksi data agentic otomatis.

Mengekstrak DOM menjadi Markdown Ramah LLM

Gunakan Crawl4AI v4 untuk ekstraksi async berperforma tinggi dan Firecrawl untuk konversi siap LLM yang mulus. (Lihat contoh kode di bagian bahasa Inggris).

Bypass Anti-Bot & Shadow DOM

Bypass Anti-Bot (Cloudflare Turnstile, Datadome):
- Proxy Residensial: Rotasi IP residensial berkualitas tinggi untuk menghindari pemblokiran IP datacenter.
- Browser Fingerprinting: Gunakan alat seperti playwright-stealth atau browser stealth khusus untuk menyembunyikan sidik jari otomatis.
- Interaksi Mirip Manusia: Tambahkan penundaan acak, simulasikan gerakan mouse yang realistis.
Shadow DOM Dinamis:
- Gunakan selektor penembus CSS atau eksekusi JavaScript untuk menembus Shadow Root.
- Telusuri pohon DOM secara rekursif dengan menyuntikkan skrip untuk mengekstrak konten.

Loop Ekstraksi Agentic

Implementasikan loop otonom di mana LLM memandu scraper berdasarkan status halaman saat ini, bukan bergantung pada selektor CSS yang rentan rusak.

Observasi: Scraper mengekstrak DOM saat ini menjadi Markdown yang bersih.
Analisis: LLM menganalisis Markdown untuk mengidentifikasi data yang diperlukan atau langkah interaksi selanjutnya (misal: "Klik tombol 'Muat Lebih Banyak'").
Aksi: LLM mengeluarkan perintah (ekstrak data, navigasi, klik, isi form).
Loop: Ulangi hingga tujuan ekstraksi tercapai.

Checklist Scraping Etis

Periksa robots.txt dan hormati aturan Disallow.
Implementasikan rate limiting.
Gunakan header User-Agent yang deskriptif.
Jangan scraping data pribadi/privat tanpa izin.

Matriks Eksekusi Berdaulat, Efisiensi Token & Invarian

Dimensi	Mandat Berdaulat & Kontrak Deterministik
Kaskade Model Frontier	Medium / High (`GPT-5.6 Sol Medium`, `Gemini 4 Argon`, `Claude Sonnet 5.5 High`, `Grok 4.7 High Fast`)
Invarian Fallback Ketersediaan	`Gemini 4 Argon/Pro` -> `Gemini 3.8 Flash`; `Claude Opus 5.5` -> `Claude Opus 4.6` (Bebas gagal saat model belum tersedia/kuota penuh)
Protokol Hemat Token & KV-Cache	Kunci instruksi & skema statis di prefix awal (`kv-cache-prefix-optimizer`); kirim potongan `CONTEXT_MAP.md` (<250 token/panggilan) saat Fan-Out
Eksekusi SaaT CLI & MCP	CLI: `vibes skill exec web-scraper <audit\|eval\|verify\|generate\|blueprint>` / MCP: `vibes_skill_dispatch`, `vibes_ast_query`, `vibes_compile_cache`, `vibes_optimize_budget`, `vibes_verify_gate`
Invarian NFR & Anti-Slop	Latensi P99 API < 85ms, paritas skema ketat OpenAPI 3.1 / TypeSpec, kontrak error RFC 9457 Problem Details, penegakan Idempotency-Key; kode 100% tuntas tanpa placeholder diverifikasi melalui `node scripts/check-anti-slop.js --strict`

web-scraper

SKILL.mdskills/web-scraper/SKILL.md