Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.xpander.ai/llms.txt

Use this file to discover all available pages before exploring further.

The gap between “I found a bug” and “engineering can fix it” is almost always a documentation problem. This agent closes that gap - structured report, verified reproduction steps, Notion page created, the moment you describe it.

​
Tutorial Summary

  • Goal: Eliminate inconsistent bug reports - every bug your team finds gets documented the same way, with the reproduction steps, screenshots, and details engineering needs to act on it immediately.
  • Estimated Time: 20–30 minutes to set up
  • What you’ll build: A conversational bug reporting agent connected to Notion and Browser Use - describe a bug, and the agent structures it, verifies it in a live browser session if needed, and creates a fully populated Notion page ready for engineering.
Every team has the same problem: bugs get described in Slack, half-logged in a ticket, or not reported at all because filing a proper report takes too long. The ones that do get filed are inconsistent - missing reproduction steps, no screenshots, vague descriptions that leave the engineering team guessing. This agent fixes that. Describe the bug conversationally, and the agent structures it, verifies it if needed, confirms it with you, and creates a clean Notion page - in the same format, every time. No forms. No manual formatting. No follow-up questions from engineering. And this is just the beginning. Every report this agent creates is structured precisely enough that a future PR agent can read it, understand exactly what’s broken, and open a fix - without a human having to translate the report into something actionable.

​
Key Features

  • Conversational input - Describe the bug the way you’d describe it to a colleague. The agent extracts what it needs and structures it automatically. No forms, no required fields, no formatting rules to follow.
  • Browser verification - The agent can open a live browser session, navigate to the relevant URL, reproduce the bug itself, capture screenshots, and use what it observed as the reproduction steps. If it can’t reproduce it, it says so - it never fabricates steps.
  • Structured output, every time - Every report follows the same five-section format: Summary, Expected Behavior, Actual Behavior, Steps to Reproduce, and Evidence. Engineering always gets what they need to act immediately.
  • Notion integration out of the box - The agent creates a fully populated Notion page with all properties set - Priority, Type, Assignee - and the full report body written as rich content blocks. No copy-pasting, no manual property filling.

​
Prerequisites

  • Account in xpander.ai
  • Notion workspace - with a bug tracking database. Connect via OAuth in the xpander.ai connectors panel and it’s ready immediately
  • Browser Use (MCP) - for live browser verification

​
Step-by-Step Implementation

Step 1 - Create the Agent and Connect Your Tools In the xpander.ai builder, create a new agent and head to the Tools tab. Attach the Notion connection - it comes pre-configured and is ready the moment you authenticate. For Browser Use, head to the MCP section and add the Browser Use remote MCP URL to connect it as an MCP server.
  • Name: AI QA Engineer
  • Model: Claude Sonnet, GPT-4o, Gemini 1.5 Pro, or any other model of your choice
Image
Step 2 - Define the Agent’s Instructions Role: 
You are an AI Bug Triage Agent. Your job is to take a conversational
bug description and turn it into a clean, structured Notion bug report
- ready for the engineering team to act on immediately.
Goal: 
For every bug reported, collect the necessary details, structure them
into a consistent five-section format, and create a fully populated Notion page
in the bug tracking database.

Never fabricate reproduction steps. If you cannot reproduce the bug, document what was observed instead.
Instructions: 
For every bug report:

1. Collect the bug details from the reporter's description. Extract:
   - What happened (actual behavior)
   - What should have happened (expected behavior)
   - Where it happened (page, URL, feature area)
   - Steps to reproduce
   - Any error messages, screenshots, or evidence provided

2. If the reporter requests browser verification, launch a live browser
   session, navigate to the relevant URL, reproduce the bug, capture
   screenshots, and use the observed steps verbatim in the report.

3. Structure the report into five sections:
   - Summary
   - Expected Behavior
   - Actual Behavior
   - Steps to Reproduce
   - Screenshot / Evidence

4. Present the full report to the reporter for confirmation within the agent before
   submitting. Allow them to approve, request changes, or cancel.

5. Once confirmed, create a Notion page in the bug tracking database
   with all properties set:
   - Priority: Now (unless specified otherwise)
   - Type: Bug (unless specified otherwise)
   - Assignee: Moriel (unless specified otherwise)

Any default can be overridden - just mention the priority,
assignee, or type when describing the bug and the agent will apply it.
Image
Once you’re happy with the instructions, hit Save and Publish. Step 3 - Add the Agent to a Channel Head to the Channels tab on the agent and connect it to wherever your team reports bugs -Slack, a chat interface, or directly via the xpander assistant. Once connected, the agent is live and ready to receive bug reports from anyone on the team.
Image
Image

​
Agent in Action

​
Scenario 1 - Basic Bug Report

Reporter: The Export to CSV button on the reports page doesn’t do anything when clicked. I’m on Chrome, logged in as an admin. Agent: Got it. Here’s the report - does this look right? 
Summary: "Export to CSV" button is unresponsive on the Reports page
Expected Behavior: Clicking the button triggers a CSV file download.
Actual Behavior: Nothing happens. No download, no error message.
Steps to Reproduce:
  1. Log in as an admin.
  2. Navigate to the Reports page.
  3. Click "Export to CSV".
  4. Observe: no file download is initiated.
Priority: Now | Type: Bug | Assignee: Moriel
Reporter: Looks good, go ahead. Agent: Bug report created in Notion. ↗

​
Scenario 2 - Browser-Verified Report

Reporter: The search bar on the dashboard returns no results even for exact matches. Please verify this in the browser. Agent: On it - launching a browser session to observe the behavior directly. [Runs live browser session…] Agent: Session complete. Here’s what was observed: 
Steps to Reproduce:
  1. Navigate to app.example.com/dashboard
  2. Click the search bar in the top navigation.
  3. Type an exact match query: "Project Alpha"
  4. Press Enter.
  5. Observe: "No results found" - despite "Project Alpha" existing.
  6. Reload the page - same result persists.
Evidence: [Live session replay URL attached]
Ready to submit? Here’s the full report for your review.

​
What the Engineering Team Sees

A Notion page lands in the bug database - fully populated, consistently structured, with reproduction steps already verified and screenshots attached. No back-and-forth asking for more context. No vague descriptions to interpret. Just a report that’s ready to act on. This isn’t limited to Notion. The same agent can push reports to Jira, Linear, GitHub Issues, or any other platform your engineering team uses to track and manage bugs - the structure stays the same, only the destination changes. This is also the foundation for what comes next. Every report this agent creates is structured precisely enough that a PR agent can read, identify the broken behavior, and open a fix automatically. The human’s role shrinks to confirming the report and reviewing the PR. What starts as automated bug logging becomes the first step in a fully automated bug-to-fix pipeline.
Image
Image