Explore your app
Start an exploration of your app, say what to check and what to leave alone, and approve the plan the subagents follow.
An exploration is how Rehearsal tests your app. You say what to check and what to leave alone, Re:agent goes through the app and writes a plan, and once you press Start testing, subagents work through that plan, each in a browser of its own on Rehearsal's servers. They record every test case they check and every bug they find, and the tests you keep afterwards are made from what they walked.
This page covers how to start one, the plan, how long it runs and what the subagents may do in your app. Watch an exploration covers the panel you follow it in, and Quick start walks through a first one from sign-up to a first run.
Start during setup Link to Start during setup
A new account ends its setup with an exploration. After your name and your workspace, setup asks one question per screen, and your answers become what Re:agent is told, so the chat does not ask them again.
Step 1: Enter the address
Type your app's address, such as
parcel.example. Rehearsal opens it once from its own servers to check that it loads. If the page links to other sites of yours, such as a marketing site beside the app, setup then asks Which one should we test?Step 2: Choose what to check first
What should we check first? offers Signing up and signing in, Getting started after signing up and The main features, which you can combine, or The whole app on its own. The last row, Your own words, takes anything else, such as "Saved cards and checkout".
Step 3: Choose what to leave alone
What should we leave alone? offers Settings and account pages and Checkout and payments, or Nothing, go ahead, and again a row for your own words.
Step 4: Start exploring
The last screen shows what Re:agent is told, under Check first and Leave alone. Press Start exploring, and the chat opens on the exploration.
An app on your computer
Enter its local address, such as localhost:3000. Setup then offers Connect with the CLI and waits until the connection is open, as Test an app on your computer explains.
Start from the chat Link to Start from the chat
You can explore again at any time from a chat in Re:agent. A new chat shows a row under the message box, Explore Parcel to find bugs, or Explore Parcel again once the app has been explored. You can also type / in the empty message box and pick Explore, which puts an Explore chip in the box, then write what to focus on or leave the box empty for the whole app. Asking in your own words works too, because Re:agent can start an exploration itself.
Re:agent then asks setup's questions in the chat, but only the ones your message and the app's last exploration leave open. If your message already says what to check, that question is skipped. If it does not, Re:agent asks What should we check first? Once it has explored the app, the question offers up to four of the app's parts, then The whole app. The part you pick becomes the focus. Re:agent asks again what to leave alone only when the focus is new. When both questions are open, they come on one card and your answers are sent together.
To get only an accessibility report, ask for it in your own words, such as "Give me an accessibility report of Parcel". Re:agent goes through the app, asks for a test account if it meets a sign-in page, and writes the report. There is no plan to approve, and nothing else is tested.
A script or a coding agent starts one with rehearsal explore or the explore tool. The command asks in the terminal for what it still needs, such as which app and what to focus on. Each question has a flag that answers it.
Started that way, nobody approves a plan or presses Start testing, so the work starts on its own. The exploration belongs to the project, so everyone who can read the project can see it. Find and fix bugs with your coding agent shows that loop from start to finish.
What to check Link to What to check
What you ask to check becomes the focus of the plan. The whole app sends the subagents through every part Re:agent finds, while a focus such as "Saved cards and checkout" keeps the plan to those parts. Write it the way you would tell a colleague. Pages are checked on a desktop screen and on a phone's, unless your words name other screens, such as "on a tablet".
Checking sign-up means making an account in your app, so the subagents may sign up with a new address from the test inbox only when you ask for it. In setup, the rows decide: picking Signing up and signing in or Getting started after signing up allows it, while your own words on that screen do not. In the chat, a focus whose words mention signing up allows it too. Otherwise the subagents use the test accounts the project already has, or stay signed out.
What to leave alone Link to What to leave alone
What to leave alone is kept in your own words, such as "The admin pages" or "Deleting a team". Re:agent leaves out of the plan every part your words name, and every subagent reads them before it starts work. The plan repeats them under What the plan won't do, so you can check them before anything starts.
Rules in your own words Link to Rules in your own words
Some things only you know, such as "A new account gets a welcome email" or "A saved card stays after signing out". While the plan waits, Re:agent may ask up to three questions about how your app should behave, each with the line Your answer becomes a rule the agents follow. under it. Skip all sets them aside, and pressing Start testing settles any that are still waiting.
Once the subagents are at work, write a new rule in the chat as you would say it, such as "Leave the billing page alone from now on". Re:agent passes your words on, and each subagent reads them when it next starts on a part, so one in the middle of a part may finish it first. From a terminal, --rule and --avoid do the same on rehearsal explore and on rehearsal explore change.
The plan Link to The plan
Before any subagent starts, Re:agent goes through your app on its own, signed out and as each test account it was given. The chat says Going through the app and planning and counts the pages it opens, with a picture of each. If it meets a sign-in page and has no account to use, it asks Can you add a test account for Parcel? When the project has accounts for that address, it asks Which test account can I use to sign in to Parcel? instead. The line under the question says what is left without one: Without an account, the agents can only test public pages. If every page it opened was a sign-in page, the line says Without an account, the agents can only test the sign-in page. Under it are pictures of the sign-in and sign-up pages it opened, each with its address. Add an account, pick one, or press Go on without an account. It never signs in through another company, such as Google.
Then the plan appears in the chat as one card, Exploration plan. It has a row for each part of your app, such as Sign up, Saved cards, Checkout and Team invitations, with how many checks it holds and on how many pages. Press a row to read its checks in the order a subagent will follow them, and use its switch to turn the part off.
Two lists under the parts say what the work will change and what it will never do. What the plan adds to your site names what the parts that are on will create, such as a saved card. What the plan won't do names what no subagent will do, such as paying for anything, messaging or inviting real people or leaving your site, and then what you asked to leave alone, in your words.
To change the plan, press Change the plan and pick Turn off a part, Add something or Ask about a part. Each starts a sentence in the message box for you to finish, or you can write the change yourself, such as "Also test inviting a teammate". Re:agent changes the card and marks the parts it changed, and the foot of the card says about how long the parts that are on will take.
Nothing starts until you press Start testing, and the plan waits for that as long as you take. While Re:agent is writing a reply, the button waits too, so a change you asked for lands first. If you leave in the meantime, Your test plan is ready waits under the bell, and the chat's row in the list of chats says Waiting for you.
How long it runs Link to How long it runs
An exploration has 30 minutes, counted from the moment you press Start testing. Subagents take the parts that are on in the plan's order, and when one finishes its part, the next part starts. How many work at once depends on how many browsers your workspace has, which is five on the Free plan, and once testing has started, the foot of the list of parts in the chat says that number.
When a check fails, a second subagent usually repeats it in a fresh browser before the failure counts as a bug, as Read a bug explains. Those second looks still run after the 30 minutes, so every failure gets one. Apart from them, nothing new starts once the time is used. A subagent still at work is told its time is up and finishes on its own, and five minutes later any subagent still on a part, at work or waiting for your answer to its question, is stopped. What a stopped subagent recorded is kept, and its part is marked Blocked on the plan. A part nobody reached in time is marked Not reached.
The exploration ends sooner when the subagents have nothing new left to check or when it reaches its spending limit, and you can stop it whenever you like, as Stop or pause shows.
The spending limit is the most one exploration may spend on the AI models behind its agents, which is 250 dollars for one started from setup or the chat. Exploring is free, so none of this is charged to you, and this limit is separate from the spending limit for additional runs in your billing. It works as a safety stop. Once too little of it is left for another subagent, the exploration ends, and its report follows as it does when the time runs out.
One exploration at a time Link to One exploration at a time
A workspace runs one exploration at a time, whoever started it and from wherever, so Rehearsal refuses to start a second until the first has ended or been stopped. A plan still waiting for Start testing counts as that one exploration, however long it waits. In the chat, Re:agent says an exploration is already in progress and asks whether you would rather wait or stop it.
To stop an exploration, you need to be able to read it, as well as the Manage projects and Use Re:agent permissions. An exploration started in a chat is private to the person whose chat it is, so only they can stop it: with the stop button in that chat, by asking Re:agent in any of their chats to stop it, with rehearsal explore stop, or through the stop_exploration tool a coding agent has beside pause_exploration and resume_exploration. Deleting the chat stops its exploration too. An exploration started from the CLI or by a coding agent belongs to the project, so anyone in it with those permissions can stop it.
When the exploration holding the workspace is in a teammate's chat, such as a plan they have not started yet, Rehearsal does not tell you which one it is, and neither you nor Re:agent can stop it. Ask your teammate to press Start testing or stop it.
What the subagents do in your app Link to What the subagents do in your app
The subagents use your app as a person would, in Chromium, so whatever they add is really added. An exploration started from setup or the chat lets them:
- Create and change things, such as saving a card or editing a profile.
- Sign in with the project's test accounts that work in that environment, such as
member, or with the one you pick when Re:agent asks how to get in. - Sign up with a new address from the test inbox, when you asked for signing up to be checked.
They never pay for anything, delete anything they did not make, or message or invite real people. From a terminal, rehearsal explore allows more with --allow, such as --allow delete, --allow send or --allow invite. It allows less with --forbid create or --forbid modify. Paying has no flag.
Before you explore production
What the plan adds to your site stays there after the exploration ends. If nothing should be added to your live app, give Rehearsal a staging address, or test accounts whose data you can lose.
When it ends Link to When it ends
When the exploration ends, Re:agent says so in the chat and its report follows as one card. The report counts the test cases checked and the bugs found, with the accessibility report and the suggestions, and offers to create tests from the main journeys the subagents walked. Read a bug explains each bug, and Create tests explains the tests.
There are two ways to go further. When no bugs were found, the report can offer the parts the plan left outside what you asked for, such as Explore Team invitations. After you fix something, a part the plan worked has Test again, which tests that part alone as a new exploration. Either way, the new exploration starts from what Re:agent wrote down about the app, which you can read and change, as What Re:agent knows explains.
Exploring is free, and so is creating tests from an exploration. Only runs of your tests count towards your plan.