Agentql

Web scraping and browser automation using AgentQL — query any webpage with natural language to extract structured data, interact with elements, and automate...

Security Scan

The skill's runtime instructions describe a networked Python library (AgentQL) that requires an API key and Playwright, but the registry metadata does not declare any required credentials, dependencies, or a source/homepage — this mismatch and lack of provenance is concerning.

Purpose & Capability

The SKILL.md content matches the declared purpose (web scraping/browser automation via AgentQL and Playwright). However the registry metadata lists no dependencies or env vars while the instructions clearly require a Python package and Playwright, so the declared requirements are incomplete.

Instruction Scope

Runtime instructions instruct the agent to use an AGENTQL_API_KEY and to call out to https://dev.agentql.com, use the 'agentql' Python package and Playwright, and optionally connect to a local browser over CDP (127.0.0.1:9222). The skill doc therefore requires network communication (sending page context to an external API) and potential control over a local browser session; the registry did not declare these data flows or permissions.

Install Mechanism

There is no install spec (instruction-only), which reduces disk-write risk. But the SKILL.md implies installing Python packages (agentql, playwright) which are not declared in metadata; missing dependency/install information is an integrity/usability gap.

Credentials

The SKILL.md requires AGENTQL_API_KEY and an external service (dev.agentql.com) but the skill metadata lists no required environment variables or primary credential. That omission hides that a third-party API key is needed and that page contents will be sent to an external endpoint.

Persistence & Privilege

The skill is not always-enabled and does not request elevated/persistent agent privileges. It is user-invocable and allows autonomous invocation (default), which is normal but increases blast radius if combined with other risks.

What to consider before installing

This skill's instructions require an AGENTQL_API_KEY, the 'agentql' Python package, and Playwright, but the registry metadata doesn't declare those or provide a source/homepage. Before installing or using it: 1) Ask the publisher for the source code or a trustworthy homepage (GitHub/PyPI) so you can inspect the agentql package and confirm what data is sent to the API. 2) Confirm exactly what the API receives (full page HTML, screenshots, or extracted fields) and the privacy/retention policy at dev.agentql.com. 3) Avoid connecting the skill to your regular, logged-in browser via CDP unless you understand the risks (it can access open sessions and sensitive data). 4) If you must try it, run it in an isolated environment (throwaway VM/container, throwaway browser profile) and use a limited, revocable API key. 5) If the author cannot provide provenance or code, treat the skill as untrusted.

AgentQL Skill

AgentQL lets you query webpages with natural language to extract structured data and automate browser interactions. It uses Playwright under the hood.

Setup

AgentQL requires an API key. Set it as an environment variable:

export AGENTQL_API_KEY="your-api-key"

Get a free API key at https://dev.agentql.com

Quick Start (Python)

import agentql
from playwright.sync_api import sync_playwright

with sync_playwright() as p:
    browser = p.chromium.launch(headless=True)
    page = agentql.wrap(browser.new_page())

    page.goto("https://example.com")

    # Query with natural language
    response = page.query_data("""
    {
        title
        main_heading
        description
    }
    """)

    print(response)
    browser.close()

Write and Run a Script

# Create a new script template
agentql new-script scraper.py

# Run it
python3 scraper.py

Common Patterns

Extract structured data

response = page.query_data("""
{
    products[] {
        name
        price
        rating
    }
}
""")

Interact with elements

response = page.query_elements("""
{
    search_box
    submit_button
}
""")
response.search_box.fill("query")
response.submit_button.click()

Headless scraping

browser = p.chromium.launch(headless=True)

With existing page (connected to running browser)

browser = p.chromium.connect_over_cdp("http://127.0.0.1:9222")  # Brave/Chrome
page = agentql.wrap(browser.contexts[0].pages[0])

CLI

agentql doctor          # Check setup
agentql new-script      # Generate template script

Notes

• API key required: https://dev.agentql.com (free tier available)
• Scripts run via: python3 script.py
• Can connect to the running Brave browser via CDP (port 9222)