Azure Playwright Browser Automation

Browser automation with Playwright CLI and Azure Playwright Service

This skill is the operational reference for browser tasks. The base prompt owns the non-negotiable lifecycle rules; this skill provides the concrete Playwright CLI command patterns.

Remote browser connection

1. Reuse an active browser session for follow-up browser work in the same hosted agent session whenever one is available. If no active browser is available, call create_session with no arguments.

2. As soon as create_session returns, call get_live_view_url. The live view URL will be delivered to the user automatically by the system. Do NOT output, repeat, or retype any URL yourself. Just acknowledge that the live view is available.

   Do not include the raw CDP URL in user-facing text.

3. Call run_playwright_cli with sessionId='browser' and the command:

   open about:blank
   

   The CDP URL is injected automatically by the server -- do NOT pass it. This must be a standalone handshake command. Do not combine it with target navigation, eval, snapshot, or any other browser operation.

4. Run all subsequent commands with the same sessionId (browser):

   goto https://example.com
   snapshot
   

   Use goto <url> for target navigation after the handshake; do not use open <url> for normal page navigation. If a follow-up command says the local session is not open, reconnect with command: open about:blank. Do not call create_session for this recovery.

5. Keep the browser session open after successful work so follow-up tasks can continue in the same live browser. Call close_browser_session only when the user explicitly asks to close the browser, when the session is unusable, or before replacing it with a fresh remote browser.

If the initial open about:blank command fails, do not retry repeatedly. Call close_browser_session, then call create_session again to create a fresh remote browser.

Common commands

Pass only the arguments shown below to run_playwright_cli.command; do not include playwright-cli or -s=<sessionId>.

# Navigation
open about:blank
goto https://playwright.dev
go-back
go-forward
reload

# Page state
snapshot
screenshot

# Interactions
click e3
dblclick e7
fill e5 "user@example.com"
fill e5 "search text" --submit
type "search query"
press Enter
hover e4
select e9 "option-value"
check e12
uncheck e12

# Date/calendar fields - use fill with the value format the field expects
fill e8 "2026-06-15"
fill e8 "06/15/2026"

# Tabs
tab-list
tab-new https://example.com/other
tab-select 0
tab-close

# Extraction and diagnostics
eval "document.title"
eval "JSON.stringify([...document.querySelectorAll('a')].map(a => a.href))"
console

After most commands, Playwright CLI emits page status and a snapshot. Use refs from the snapshot, such as e15, for subsequent interactions.

Cleanup

Do not close a healthy remote browser at the end of a normal task. When the user asks to close the browser, or when replacing a broken session, call close_browser_session with just sessionId='browser'.