Web Automation — E-commerce Checkout with Human Review

Automate an end-to-end e-commerce flow with Browser Agents: search for a product, add it to the cart, pause for a human approval, then complete checkout using credentials stored securely in the Vault. This guide walks the whole build on the Workflow Canvas.

What you'll build

By the end you'll have a published workflow that takes a product_name and a starting_url, drives a browser to search and cart the item, emails the cart to a reviewer, and — only on approval — finishes checkout.

What you'll learn

• Use a Webhook Trigger to pass dynamic inputs (starting_url, product_name).
• Configure a persistent Browser Session and link login credentials via Vault Management.
• Orchestrate two Browser Agents for a multi-step task (Search & Cart → Checkout).
• Insert a Human-in-the-Loop (HITL) approval step and branch on its result with a Router.
• Define Input/Output Schemas so agents pass structured data downstream.

Architecture at a glance

Before you begin

Make sure the following are in place — the first time through this takes 2–3 minutes:

• A FlowGenX account with access to Design → Flow Designer → Workflow Canvas.
• An LLM provider API key (OpenAI, Anthropic, or Gemini).
• E-commerce login and shipping credentials in the Platform Vault (you can also add these later in the Browser Session node).

---

Step 1 — Add a Webhook Trigger and define its schema

Goal: set up the workflow entry point and capture dynamic inputs. (~2 min)

1. Drag the Webhook Trigger node onto the canvas.
2. Click the node to open its configuration panel on the right.
3. Open the Schema tab and configure the request payload and response.
4. Add two required request fields: product_name (string) and starting_url (string).
5. Click Apply.

Step 2 — Add and configure a Browser Session

Goal: establish a persistent browser environment and link secure credentials. (~3 min)

1. Drag the Browser Session node onto the canvas, below the Webhook Trigger, and connect their edges — the trigger's output feeds the session.
2. Click the node to open its configuration panel.
3. Set a starting URL (e.g. www.google.com) where the session begins.
4. Under Vault credentials, select the stored login for your site (e.g. Amazon-Login-Vault), or create a new one.
5. In advanced settings, enable Persist cookies and headless mode.
6. Click Apply.

Step 3 — Configure the first Browser Agent (Search & Cart)

Goal: log in, search for the product, add it to the cart, and output the cart details. (~5 min)

1. Drag a Browser Agent node after the Browser Session and connect their edges.

2. Open the node's Configuration panel.

3. Under Models → LLM Model, choose or create a model (e.g. gpt-4o).

4. Under Prompts → Prompt, paste the system prompt:

   You are an e-commerce automation assistant. Log in with the credential provided from the Vault, search for the specified product, and add it to the shopping cart. Use the securely managed Vault credentials for login. Then output cart_info and cart_url.

5. (Optional) In Steps, add a line-by-line plan for the agent to follow.

6. On the Schema tab → Input, define two inputs — starting_url and product_name — and map each to the Webhook Trigger's output (via a JEXL expression or by clicking the field in the Context Explorer).

7. On the Schema tab → Output, define two output fields — cart_info and cart_url. These are structured fields the agent will populate; they don't need an expression.

8. Click Apply.

Step 4 — Add a Human-in-the-Loop (HITL) review

Goal: pause the workflow and email the cart to a reviewer for explicit approval before checkout. (~3 min)

1. Drag the Human-in-the-Loop node onto the canvas, downstream of the first Browser Agent.
2. Set Notification Type to Email.
3. Set the Recipient (e.g. [email protected]).
4. In the Context Builder, surface the cart details: Cart Info: {{browser_agent.output.cart_info}}

5. In the Form Builder, add a required field named proceed_to_checkout (Boolean / checkbox).

6. Check the Preview to see how the form will look to the reviewer.

7. On the Input/Output tab → Output, confirm the proceed_to_checkout field is present (it's added automatically from the Form Builder). Then click Apply.

Step 5 — Add a Router to branch on the decision

Goal: route the workflow based on the reviewer's proceed_to_checkout answer. (~2 min)

1. Drag the Router node after the HITL node and connect them.
2. Click the Router to define routes.
3. With Expression-Based routing enabled, add the expression proceed_to_checkout === true on route a. You'll connect the checkout agent to this path.
4. Leave the fallback route as-is, then click Apply.

Step 6 — Configure the final Browser Agent (Checkout)

Goal: resume the session and complete the order using Vault-stored details. (~4 min)

1. Drag a second Browser Agent onto the canvas and connect the Router's a path to it.

2. In its Configuration panel, it re-uses the same Browser Session from Step 2.

3. Under Prompts → Prompt, paste the system prompt:

   You are a final-step automation assistant. Navigate to the cart, verify the contents, and complete checkout using the pre-configured shipping and payment details from the Vault.

4. (Optional) In Steps, outline the plan: navigate to the cart URL, click Proceed to Checkout, fill shipping/payment fields from the Vault, and submit the order.

5. On the Schema tab → Input, define cart_url (mapped from Browser Agent 1 → output → cart_url) and product_name (from Webhook Trigger → output → product_name). On Output, define raw_order_data to capture the order-confirmation summary.

6. Click Apply.

Step 7 — Save and publish

Goal: persist your work and promote it so it can be triggered. (~2 min)

1. Click Save (bottom-left of the canvas) — the workflow is saved as a draft.
2. Click Publish (or Republish on later edits).
3. Wait for the success toast: "Workflow published successfully."

Step 8 — Send a test request

Goal: fire a sample request and watch the agents reason and act. (~3 min)

1. Open the Webhook Trigger node again and switch to test mode.

2. Scroll to Deploy API and click it to expose the workflow as an API. In the Credentials section just below, attach a pre-configured API key to secure it, then click Save for Testing.

3. In the Test Request section, fill the request body to match your schema (or copy the generated cURL), then click Send Test Request.

4. After a moment you'll receive a success response.

Step 9 — Monitor the execution

Goal: review the run, trace the path, and inspect node-level inputs and outputs. (~2 min)

1. Go to Monitor → Workflow Runs.
2. Select your Web Automation workflow to see its execution history.
3. Click the run ID for the test you just sent.
4. Review the end-to-end trace — it spans every step and agent call, including node_name, node_type, and the Router's routing_path.
5. Click any node (Browser Agent, HITL, …) to see its captured Input and Output state. Failed nodes capture an error_message and status for debugging.

You're done — what's next

You've built, published, and tested an e-commerce automation with a Human-in-the-Loop review. Keep going:

• Introduction to Agents — React, Supervisor, and Swarm agent patterns.
• Human-in-the-Loop — deeper on approval steps, forms, and channels.
• Traceability — move from manual tests to full run observability.

Troubleshooting