Before you start
Open Codex, Claude Code, Cursor or another compatible coding agent in your project. The environment needs Node.js 22 or newer, Chrome or Chromium and browser tools that can connect to the browser Rill provides.
You can begin without a Rill account. Anonymous recordings upload to Rill and receive an unlisted link; anyone with that link can view the video and diagnostics. Use suitable test data. See anonymous limits and expiry.
Give your agent a task
Tell the agent which app, steps and expected behavior matter. Paste this prompt into the conversation for that project. It asks the agent to check setup and make one useful recording.
Read and follow the Rill skill at https://userill.dev/skill.md for this project. Install the required Rill tools if needed, following this environment's setup and approval rules. Record one useful browser flow for this project. Follow our current task when clear; otherwise choose a short, non-destructive smoke test. Use any specific app URL, steps, test data and expected behavior I supplied. Check prerequisites and confirm that you can control the browser Rill provides. Use the existing app setup. Ask if the target is unclear, application changes are needed, or an action needs approval. Do not submit payments or make destructive changes without approval. Start anonymously if I am not connected, following the existing skill and approval rules. Before anonymous capture, explain that the recording uploads to Rill with an unlisted link that anyone holding it can open. Use suitable test data and avoid sensitive content. Stop the recording, wait for processing, and verify that it plays. Return the link, expiry, tested version if available, what you checked, what you observed, and anything you could not verify. Report failed checks or processing problems explicitly.
Read the Rill skillFor example, add: “Use our running local app. Open the product list and filter by category. Check that only products in that category remain.”
What happens next
- The agent reads the skill and checks the CLI, browser, encoding and its ability to control the recorded browser.
- It uses your current task or chooses a short, non-destructive smoke test. It asks if the target is unclear or an action needs approval.
- It starts the existing app setup, starts a Rill recording and connects its browser tools to that recording.
- It performs the steps, stops the recording and waits for playback to be ready.
- It returns the link, observations, limits of the check and expiry.
Your agent operates the browser; Rill captures it. No Chrome extension is required for this local-agent workflow. Localhost must be reachable from the agent’s environment.
Check the handoff
Open the returned link and play it. The handoff should say what the agent checked, what happened and what it could not verify.
Recording: <your playable Rill link>
Checked: Opened the product page and filtered the list.
Observed: The selected filter changed the visible products.
Not checked: Checkout, payment, other filters and other browsers.
Tested version: <commit or build, if known>
Expires: <the expiry returned by Rill>This is example output, not a recording. A link alone is not proof of success. The agent should report processing failures and incomplete checks. Use agent troubleshooting if it cannot finish.
Keep the recording
rill connect --email you@example.comOpen the link printed by the CLI, sign in with that email, enter the agent’s code and approve. Your unexpired anonymous recordings move into the account and keep their existing share URLs. Finish or cancel active recordings first.
Connecting is not a paid upgrade and does not make those links private. Review the workspace access and renewal permissions before approving.