Browser Automation

Type ID: core.browserAutomation  ·  Kind: Action  ·  Trace: TOOL

Launches a Chromium browser with Playwright, optionally performs a form login (with TOTP support), then runs an ordered list of steps — navigate, fill, click, wait, and extract — returning the collected data.

Credentials

None

Properties

Property	Key	Type	Required	Default	Possible values	Shown when
URL	url	string	Yes	''	Start URL (or metrics URL)	Always
Timeout (ms)	timeoutMs	number	No	30000	Default page timeout in milliseconds	Always
Include HTML	includeHtml	boolean	No	false	Include page HTML in output (can be large)	Always
Auth Type	auth.type	options	No	none	none — None; form — Form Login	Always
Login URL	auth.loginUrl	string	No	''	URL of the login page	auth.type = form
Username Selector	auth.usernameSelector	string	No	#email	CSS selector for the username field	auth.type = form
Password Selector	auth.passwordSelector	string	No	#password	CSS selector for the password field	auth.type = form
Submit Selector	auth.submitSelector	string	No	button[type=submit]	CSS selector for the submit button	auth.type = form
OTP Selector (optional)	auth.otpSelector	string	No	''	Selector for the OTP input, if the site asks for one	auth.type = form
OTP Submit Selector (optional)	auth.otpSubmitSelector	string	No	''	Selector to submit the OTP	auth.type = form
Use TOTP (Authenticator) Automatically	auth.useTotp	boolean	No	false	Generate the OTP from a TOTP secret in credentials	auth.type = form
Steps	steps	array	No	[]	Ordered automation steps (see below)	Always

Steps (nested fields)

Each array entry is a step object with a type and type-specific fields.

Field	Key	Type	Default	Possible values	Shown when
Type	type	options	goto	goto — Go To; fill — Fill; click — Click; waitForSelector — Wait For Selector; waitForTimeout — Wait For Timeout; extract — Extract	Always
URL	url	string	''	Target URL	type = goto
Wait Until	waitUntil	options	networkidle	load — Load; domcontentloaded — DOM Loaded; networkidle — Network Idle	type = goto
Selector	selector	string	''	CSS selector	type = fill, click, waitForSelector
Value	value	string	''	Value to fill	type = fill
Timeout (ms)	timeoutMs	number	15000	Wait timeout in milliseconds	type = waitForSelector, waitForTimeout
Selectors	selectors	array	[]	Extraction rules (see below)	type = extract

Extract selectors (nested rule fields)

Each entry is a selectorRule object:

Field	Key	Type	Default	Possible values	Shown when
Key	key	string	metric	Output key for the extracted value	Always
CSS Selector	selector	string	[data-testid=metric]	CSS selector to read from	Always
Type	type	options	text	text — Text; texts — Texts; attr — Attribute; attrs — Attributes	Always
Attribute Name	attrName	string	href	Attribute to read	type = attr, attrs

Notes

• Parameters are resolved through the resolver engine before execution.
• The browser is launched with chromium.launch({ headless: false }) — a non-headless Chromium — using a fresh newContext/newPage. The page's default timeout is set from timeoutMs.
• Form login (auth.type = form) navigates to loginUrl (networkidle), fills username and password, clicks submit. If otpSelector is set it waits up to 8000 ms for it; on timeout it returns silently. When the OTP field appears and useTotp is true, a 6-digit TOTP is generated from auth.totpSecret (via otplib), filled in, optionally submitted via otpSubmitSelector, then waits for networkidle. If the OTP field is present but useTotp is false, it throws OTP required but useTotp is disabled.; if useTotp is true without a secret it throws.
• Steps run in array order: goto navigates to step.url (falling back to the top-level url) with the chosen waitUntil; fill and click act on the selector; waitForSelector waits for the element; waitForTimeout sleeps. extract reads each rule — text → first element's textContent, texts → all text contents, attr → first element's attribute, attrs → that attribute across all matches — into the extracted map under rule.key.
• After each step, if includeHtml is true the current page.content() is captured as lastHtml.
• Output is a single item { url, finalUrl, title, extracted, html?, timings: { totalMs } }. Errors are caught and returned as { error: true, message, timings } (not thrown). The page and browser are always closed in finally.