Build an Agent

Agents are multi-step automations built visually in the Cloud UI. You drag blocks onto a canvas, wire them together, configure each one, and run the whole sequence with a click. No code required. Agents can be versioned, shared across your team, scheduled on a cron, and run with different parameters each time. If you prefer writing automation code in Python or TypeScript, see Browser Automation in the developer docs. Both approaches handle multi-step work across pages. Agents are the visual path, browser automations are the code path.

Agents vs. Browser Automation: Agents are defined in this visual editor and run from the Cloud UI (or triggered from code). Browser automations are written in Python/TypeScript using Page, Agent, and Browser. Choose based on whether your team prefers a visual editor or a code editor.

This page explains how agents work, walks you through building one from scratch, and covers the editor’s features.

How agents work

An agent is a sequence of blocks that run one after another. Each block does one thing: navigate to a page, extract data, send an email, run custom code, loop, or branch. You connect blocks on a visual canvas, pass data between them, and run the whole chain with one click.

[Browser Task] → [Extraction] → [Loop] → [Send Email]
       ↓               ↓            ↓           ↓
  opens the page   JSON output  per-item     digest sent
                                processing

How data flows between blocks

Every block produces an output. Downstream blocks reference it using the pattern {{ block_label_output }}. For example, an Extraction block labeled scrape_listings produces output that later blocks access as {{ scrape_listings_output }}. The label is lowercased with spaces replaced by underscores, then _output is appended. This is how you chain steps together. One block’s output becomes the next block’s input.

Block categories

Category	What it does	Examples
Browser Automation	Interact with web pages using AI	Browser Task, Browser Action, Extraction, Login, Go to URL, Print Page
Data and Extraction	Transform data without a browser	Text Prompt, File Parser
Control Flow	Branch, loop, validate, or pause	Loop, While Loop, Conditional, AI Validation, Code, Wait
Files	Download and upload files	File Download, Cloud Storage Upload
Communication	Send data to external services or humans	Send Email, HTTP Request, Human Interaction

See Block Reference for every block type and its configuration fields.

If your block needs a browser, pick a browser automation block. If it just needs to process data, use Text Prompt or Code. If you need to repeat or branch, use control flow.

Tutorial: Build a job listing tracker

Let’s build a real agent. You’ll create an automation that finds the top 5 job listings from Hacker News’s monthly “Who is Hiring” thread, summarizes each one, and emails you a digest. This uses 5 blocks across 3 categories: browser automation, data and extraction, and communication.

Create a new agent

Go to Agents in the left sidebar and click Create → Blank Agent.

Create agent dropdown showing Blank Agent and From Template options

The editor opens with an empty canvas showing a single start node. Click the agent title at the top left and rename it to HN Job Tracker.

Add a Browser Task block

Hover over the + button on the edge below the start node and select Add Block. A searchable block library opens on the right. Find and select Browser Task Block.The block appears on the canvas with a configuration panel on the right. Fill in:

Field	Value
Label	`find_hiring_thread`
URL	`https://news.ycombinator.com`
Prompt	`Find and open the latest monthly "Ask HN: Who is hiring?" thread. The task is complete when the thread is open and job listing comments are visible.`

The Browser Task block uses AI to navigate the page and accomplish the goal described in your prompt. Include your success criteria directly in the prompt so the AI knows when it’s done.

Browser Task block configured with find_hiring_thread label, URL, and prompt

Why Browser Task instead of Go to URL?

The URL for the latest hiring thread changes every month. The Browser Task block uses AI to find the right link on the page. Use Go to URL when you know the exact destination URL upfront.

Add an Extraction block

Hover over the + on the edge after find_hiring_thread, select Add Block, then choose Extraction Block.Configure:

Field	Value
Label	`scrape_listings`
Data Extraction Goal	`Extract the top 5 job listings. For each listing, get the company name, role title, and a brief description of the position.`

Now enable the Data Schema checkbox to reveal the JSON editor, and paste:

{
  "type": "object",
  "properties": {
    "listings": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "company": { "type": "string" },
          "role": { "type": "string" },
          "description": { "type": "string" }
        }
      }
    }
  }
}

The schema tells Skyvern exactly what structure to return. Without it, the AI picks its own format. You can also click Generate with AI next to the schema field to have Skyvern suggest a schema based on your extraction goal.

Extraction block configured with scrape_listings label, extraction goal, Data Schema checkbox enabled, and JSON schema

Label your blocks descriptively. The label determines the output variable name. scrape_listings means downstream blocks reference its data as {{ scrape_listings_output }}.

Add a Loop with a Text Prompt inside

Hover over the + after scrape_listings, select Add Block, then choose Loop Block.Configure the loop:

Field	Value
Label	`process_each`
Loop Value	`{{ scrape_listings_output.extracted_information.listings }}`

This iterates over the 5 listings extracted in the previous step. Now add a block inside the loop. Hover over the + inside the loop body, select Add Block, and choose Text Prompt Block.Configure the Text Prompt:

Field	Value
Label	`summarize`
Prompt	`Summarize this job listing in one sentence. Company: {{ current_value.company }}, Role: {{ current_value.role }}, Description: {{ current_value.description }}`

Enable the Data Schema checkbox and paste:

{
  "type": "object",
  "properties": {
    "summary": { "type": "string" }
  }
}

The Text Prompt block sends text to the LLM without opening a browser. It processes data only. Each iteration produces a one-line summary of that job listing.

Loop block with process_each label containing a summarize Text Prompt block inside

What are current_value and current_index?

Inside a loop, two reserved variables are available:

{{ current_value }}: the current item from the list being iterated
{{ current_index }}: the item’s position, starting from 0

These only exist inside loop blocks. In this example, {{ current_value.company }} gives you the company name for the current listing.

Add a Send Email block

Hover over the + on the edge after the loop (back on the main flow, not inside it), select Add Block, then choose Send Email Block.Configure:

Field	Value
Label	`send_digest`
Recipients	Your email address
Subject	`HN Job Digest - Top 5 Listings`
Body	`Here are today's top job listings from Hacker News:\n\n{{ process_each_output }}`

The {{ process_each_output }} reference pulls in the collected summaries from every loop iteration.

Save and run

Agents are not auto-saved. Click the Save button (disk icon) in the header bar before navigating away or you’ll lose your work.

Click Save, then click Run in the top right. A parameters page opens where you can review and confirm the agent’s settings. Click Run Agent to start execution.The live execution view opens. Watch as Skyvern navigates Hacker News, extracts listings, summarizes each one, and sends your email.

What you built

Five blocks, three categories, zero code. You can now re-run this agent anytime, tweak the extraction schema to pull different fields, swap the email for an HTTP Request to post to Slack, or add a Conditional block to only email when listings match certain criteria.

Editor reference

Layout

Area	What it shows
Header bar (top)	Agent title (click to rename), Parameters button, Save button, Template toggle, History button, Run button
Canvas (center)	Visual graph of blocks connected in sequence. Zoom with the scroll wheel, pan by dragging the background.
Configuration panel (right)	Opens when you click a block. Shows that block’s settings form.

Adding blocks

Hover over the + button on any edge (the connecting line between blocks) to reveal the menu. Select Add Block to open the block library, a searchable list of all available block types. Click a block to insert it at that position. The menu also offers Record Browser (record actions in a live browser to generate blocks) and Upload SOP (upload a PDF procedure to auto-generate an agent). When you create a blank agent, hover over the + on the start edge to add your first block.

Configuring blocks

Click any block on the canvas to open its configuration panel on the right. Each block has a Label field (which determines its output variable name) and block-specific settings. Browser-based blocks share additional fields like URL, Engine, and Max Retries. See Block Reference for the full list of fields on each block type.

Connecting and reordering blocks

Blocks auto-connect when you insert them via the + button. They execute top to bottom in the order shown on the canvas. Conditional blocks create branches. Each branch connects to a different sequence of blocks and merges back at a common point. Loop blocks contain child blocks that repeat for each item in the loop’s list. While Loop blocks contain child blocks that repeat while their condition remains true. To reorder blocks, delete the block you want to move and re-add it at the desired position using the + button.

Deleting blocks

Click the three-dot menu on a block’s header and select Delete Block. Downstream blocks reconnect to the previous block automatically.

Saving

Agents are not auto-saved. Click the Save button (disk icon) in the header bar to persist your changes. The editor warns you if you try to navigate away with unsaved work.

The Run button also saves before starting execution.

Header bar actions

Button	Description
Save (disk icon)	Save the current agent definition
Template (bookmark icon)	Toggle whether this agent is saved as an organization template
History (clock icon)	View past runs of this agent
Schedule (clock icon)	Create or manage recurring schedules for this agent (see Create a Schedule)
Parameters	Open the parameters panel (see Agent Parameters)
Run (play icon)	Save and start a new execution (see Running Agents)

What’s next

Block Reference

Configuration fields for every block type

Add Parameters

Pass dynamic values into your agents

Run Your Agent

Execute agents and track results

Schedule This Agent

Set up recurring cron-based execution

Getting Started

Building Agents

Agent Blocks

Running Tasks

Re-using Browsers

Viewing Results

Managing Credentials

Integrations

Account & Settings

How agents work

How data flows between blocks

Block categories

Tutorial: Build a job listing tracker

What you built

Editor reference

Layout

Adding blocks

Configuring blocks

Connecting and reordering blocks

Deleting blocks

Saving

Header bar actions

What’s next

Block Reference

Add Parameters

Run Your Agent

Schedule This Agent

​How agents work

​How data flows between blocks

​Block categories

​Tutorial: Build a job listing tracker

​What you built

​Editor reference

​Layout

​Adding blocks

​Configuring blocks

​Connecting and reordering blocks

​Deleting blocks

​Saving

​Header bar actions

​What’s next

Block Reference

Add Parameters

Run Your Agent

Schedule This Agent

How agents work

How data flows between blocks

Block categories

Tutorial: Build a job listing tracker

What you built

Editor reference

Layout

Adding blocks

Configuring blocks

Connecting and reordering blocks

Deleting blocks

Saving

Header bar actions

What’s next