Building Flows - Firstwork

Agents are built using a visual instruction editor. The editor presents a canvas where you compose a sequence of instructions that the agent follows step by step.

Instruction Types

Instruction	Description
Visit URL	Navigate the browser to a specific address
Click Element	Click on a page element identified by a selector
Type Input	Type text into a form field
Upload File	Upload a file to a designated element
Download File	Download a file from a URL
Enter Iframe	Switch the agent’s context into an embedded iframe
Take Screenshot	Capture a screenshot of the current page
Wait for Element	Pause until a specific element appears on the page
Wait for URL	Pause until the browser URL contains a specific pattern
Solve CAPTCHA	Attempt to solve a CAPTCHA challenge
AI Instruction	Give the agent a natural-language instruction and let AI determine the actions
Run JavaScript	Execute custom JavaScript code in the browser
Fetch OTP	Retrieve a one-time password from a configured source
Loop Over Data	Iterate over a data array, executing child instructions for each item
Conditional Branch	Evaluate a condition and follow the TRUE or FALSE path
Terminate	End the agent’s execution

Element Selection

When an instruction targets a page element, the agent identifies it using one of these methods:

Method	Description
CSS Selector	Standard CSS selector syntax
XPath	XPath expression
ID	Element’s HTML ID attribute
Class Name	Element’s CSS class
Tag Name	HTML tag name
Link Text	Full text of a hyperlink
Partial Link Text	Partial match on hyperlink text
Text Containing	Match by text content

How Agents Are Triggered

Method	Description
Automation Action	The “Run AI Agent” automation action triggers execution with application context
Form Step	An “AI Agent” step in a hiring flow triggers the agent when the candidate reaches it
Manual Trigger	Admins can trigger an agent directly, selecting a hiring flow and the applications to process
Test	Run the agent from the editor with streaming output and debug mode

Testing

The test interface allows you to run the agent with:

Streaming mode — Watch the execution output in real-time
Debug mode — Verbose logging for troubleshooting
Input variable values — Provide text or file values for each input
Loop data — For agents with loop instructions, provide the data array

Execution History

Every agent execution creates a record that tracks:

Execution status and timing
Input data provided
Output data captured
Detailed execution logs