Rill
← All articles

Guides

Browser checks with parallel AI agents

The Rill Team

You have two coding agents working on one app. One is fixing checkout, the other onboarding. Each has its own worktree. You ask the checkout agent to test a cancelled payment, and seconds later the browser is on the welcome screen. The agent cannot find the payment button and reports a failure.

From its summary alone, you cannot tell whether the app redirected or another agent took over. Give each agent its own recorded browser with Rill, then watch the steps behind the result.

1. Label the feature you are checking

Give the payment work a distinctive name, for example Project Orion, and choose another for onboarding, such as Project Soren.

Ask each agent to add a small badge beside the app logo in its development preview:

  • Orion · Payments
  • Soren · Onboarding

Set the badge from local configuration, keep it out of production and check it matches the running preview. Use the same name in the task and recording title. When you switch windows or open a Rill recording, you can see which change you are reviewing.

2. Start a separate Rill browser for each agent

Install the CLI in the environment where your agent works:

npm install --global @runalabs/rill-cli@latest

Give Codex, Claude Code, Cursor or another compatible agent the Rill skill. It walks through prerequisites, browser attachment and returning a playable recording.

By default, a local rill record start launches a new managed browser with its own profile. Start one recording for each agent, with the correct preview URL and a useful title. For example:

# Payment agent
rill record start --browser new --url <payment-preview-url> --title "Orion: cancel checkout"

# Onboarding agent
rill record start --browser new --url <onboarding-preview-url> --title "Soren: first useful recording"

Each agent must connect its browser tools to the cdpUrl returned by its own start command. Orion then drives the payment browser while Soren drives the onboarding browser. Both can return a Rill link for you to review.

3. Give each agent one question to answer

For Orion, ask whether cancelling checkout returns the user to the intended page with their selection intact. Start the preview and confirm the payment provider’s test environment before recording.

Read and follow https://userill.dev/skill.md.
Record the Orion checkout-cancellation flow at [preview URL] using a new managed browser.
Confirm the feature badge and designated test account. Use the payment provider’s test environment.
Open checkout, cancel and check [expected destination or selection]. Stop before any purchase.
Stop the recording, verify playback and return the link, observed result and tested version if known.

For Soren, use a designated new test account. Follow onboarding through one short recording, processing and playback, then return to the workspace to check that the result is available. You get a walkthrough of the first useful outcome, rather than only the welcome screen.

4. Watch the result and send precise feedback

Open each Rill link and play the moment that matters. Agent recordings include captured console messages, network requests and a browser timeline alongside the video. Use those details to investigate a failed request or an unexpected navigation.

If a check was interrupted, restore the starting state and ask the agent to record the steps again in its own browser. Keep both links in your review note. If the issue recurs, you have a more focused reproduction to investigate.

Ask for a short handoff:

Feature: Orion · Payments
Checked: Opened test checkout, cancelled and inspected the return page.
Observed: [what happened]
Target: [preview URL and verified build, or version unknown]
Recording: [playable Rill link and useful timestamp]

Choose test data you are comfortable sharing, review the recording, then send the link with your feedback.

Record your next change

Give your agent the Rill skill, name the feature and ask it to record one complete flow. Open the returned link to see what it checked and give feedback on the step that needs work.

Make your first Rill recording, or see current plans to find the right fit for your workflow.