Advai Browser

Advai Browser

Use this skill to operate a live browser through advai browser.

When to Invoke

Invoke this skill when the user wants to:

• Open or navigate real web pages
• Click, type, fill, select, or scroll in a browser
• Read DOM content, extract structured data, or execute JavaScript
• Capture screenshots, inspect cookies, or manage tabs
• Scrape public web content or validate browser-side behavior

Do not use this skill for static codebase inspection or for platforms that already have a more specific native skill.

Preconditions

Before running browser actions:

1. This skill depends on the advai CLI from https://github.com/Advai-X/advai-cli.
2. Ensure advai is installed and available on PATH.
3. Ensure the advai browser subcommand is available in the current installation.
4. If advai browser is missing, stop and guide the user to install or upgrade advai from https://github.com/Advai-X/advai-cli before continuing.
5. Ensure the browser extension/runtime is installed and connected.
6. Start with:

advai browser doctor

If browser support is unavailable, surface the doctor output and explain what is missing.

Core Workflow

Use a named session for each task, for example demo, zhihu, checkout, or research.

Basic flow:

advai browser open demo https://example.com
advai browser wait demo --selector "h1"
advai browser get demo
advai browser exec demo "document.title"

Reuse the current tab instead of creating a new one:

advai browser open demo https://example.com --replace

Important behavior:

• open creates a new tab by default
• open --replace reuses the current tab when the session already has one
• If the session has no active page, open --replace falls back to opening a new tab
• Tabs are not closed automatically after the action completes

Common Commands

Open and navigate:

advai browser open <session> <url>
advai browser open <session> <url> --replace
advai browser navigate <session> <url>
advai browser state <session>

Interact with the page:

advai browser click <session> <selector>
advai browser type <session> <selector> <text>
advai browser fill <session> <selector> <text>
advai browser select <session> <selector> <option>
advai browser keys <session> <text>
advai browser scroll <session> --selector <css>
advai browser back <session>

Read and extract:

advai browser get <session> [--selector <css>]
advai browser find <session> <selector>
advai browser extract <session> [code]
advai browser exec <session> "<js>"
advai browser wait <session> --selector <css>
advai browser wait <session> --text <text>

Session utilities:

advai browser cookies <session> [--domain <domain>]
advai browser tabs list <session>
advai browser tabs new <session> --url <url>
advai browser tabs select <session> <index>
advai browser tabs close <session> <index>
advai browser close <session>

Advanced capture:

advai browser screenshot <session> --output page.png
advai browser network start <session> --pattern <text>
advai browser network read <session>
advai browser frames <session>
advai browser cdp <session> <method> --params '{"key":"value"}'

Operating Guidelines

• Prefer wait before reading or clicking dynamic content.
• Prefer get, find, extract, or exec over brittle string parsing.
• Use stable selectors whenever possible.
• Use explicit session names so multi-step tasks remain traceable.
• Clean up with tabs close or close when the workflow intentionally ends.
• For scraping tasks, first validate page readiness, then extract structured data.
• If advai browser is not recognized at runtime, do not guess alternate commands; direct the user to install or upgrade advai from https://github.com/Advai-X/advai-cli.

Examples

Search a page:

advai browser open zhihu https://www.zhihu.com/hot
advai browser wait zhihu --selector "body"
advai browser extract zhihu "Array.from(document.querySelectorAll('a')).slice(0,10).map(a => a.innerText).filter(Boolean)"

Reuse an existing session tab:

advai browser open research https://trends24.in/ --replace
advai browser wait research --selector "body"
advai browser screenshot research --output trends.png