Watch an exploration

Follow the subagents of an exploration as they work, read what each step in the chat means, and stop an exploration early.

While an exploration runs, you can follow every subagent as it works. The chat keeps one block for the exploration, with a line for each subagent and what it last did, and the panel beside the chat shows their browsers live, the test cases they check and the bugs they find. This page explains what each part shows and how to stop an exploration early.

The panel Link to The panel

Beside a chat that is exploring, the panel belongs to the exploration, and on a wide screen it opens by itself when you arrive on the chat while the exploration runs. Its top line names the app and how long the exploration has run, and under it is a list of views, each with a count once it holds something. If you closed the panel, the panel button at the top right of the chat opens it again, and the buttons at the panel's own top make it full screen or close it.

ViewWhat it holds
SubagentsEvery subagent, with how many are at work, or how many finished once the exploration has ended.
Test casesWhat the subagents check, by part of the app, with how many cases are checked of those planned.
BugsThe bugs they found, as Read a bug explains.
AccessibilityThe pages' accessibility issues, as Suggestions and accessibility explains.
SuggestionsSmall things on the pages that could be better.
RecordingsThe recording of every browser the subagents used.

Press a view to open it in the panel, and the arrow at the top to come back. Under the views, Next steps offers what to do once there is something to act on, such as connecting your coding agent to fix the bugs or creating tests from what passed.

Subagents and their browsers Link to Subagents and their browsers

Each subagent is named after what it checks, such as Saved cards, and has an icon of its own that it keeps in the chat and in the panel. Subagents lists them under Active and Done, with what each is doing now or how long ago it finished. A subagent that has no browser yet says Waiting for a browser until one is free.

A second look stands under the subagent whose failed check it repeats. It works in a fresh browser, and the failure it repeats counts as a bug only once the second look sees it too. A subagent whose work did not go to plan carries one of these marks:

MarkWhat it means
WaitingIt asked you a question and waits for your answer, as Questions from Re:agent explains.
BlockedSomething in your app stopped it, or you skipped its question, and its place says why.
Could not finishSomething went wrong on Rehearsal's side. In an exploration from setup or the chat, Rehearsal can hand its part, with the checks already made, to the next free subagent while there is time. A part that two subagents could not finish is marked Blocked on the plan.

Press a subagent to open its place. Its browsers are listed there, and the one you watch grows into a player: the live picture while the subagent works in it, then its recording once it has closed. Under the player, the steps it took are listed page by page, and pressing a step moves the recording to that moment. The steps of a bug it found are violet on the recording's bar, and the last of them carries the bug's number, such as Bug #5.

Test cases Link to Test cases

A test case is one thing a subagent checks, written as a sentence, such as "A saved card is still there after signing out". Test cases groups them by part of the app, and a big part by page, with the bugs and the number of cases checked beside each part. Every case carries a mark:

MarkWhat it means
PassedThe app did what the case expects.
BugThe app did something else.
BlockedThe subagent could not get far enough to check it.
Could not tellThe subagent could not tell from the page whether the app did what it should.
CheckingA subagent is checking it now.
Not checkedNobody has checked it.

Press a case to open it. It says what the app should do under Expected, what the subagent saw under What happened, and Watch opens the recording at the steps that checked it.

Recordings Link to Recordings

Recordings lists every browser of every subagent, with a still of its recording, how long it ran, the account it was signed in as, how many steps it took and the bugs it showed. A browser still open shows Live in place of its still. Press a row to open it in its subagent's place, where it plays with its steps. The recordings stay with the exploration, so they are still there after you reload the page.

Steps in the chat Link to Steps in the chat

The exploration's block in the chat grows while it runs and reads from top to bottom in the order things happened:

  • Opened and a count of pages is Re:agent's first pass through the app, with a picture of each page it opened.
  • The plan, once you pressed Start testing, with each part marked Waiting, Working, Done, Blocked or Not reached.
  • Started 5 subagents with their icons, each time a group of subagents starts. Press an icon to open that subagent in the panel.
  • One line for each subagent at work, which changes in place with the step it is on.
  • A line each time a subagent checks a case, cannot check one, finds a bug or asks you something, led by its name, such as Saved cards found a bug · Saved card disappears after signing in.
  • Once a subagent is done, its lines fold into one, such as Saved cards · checked 6 cases, found 1 bug. Press the line to open them again.
  • A last line that says how many subagents are working, how many cases are checked of those planned, and for how long. When the exploration ends, the block folds to Explored for and how long it took.

Re:agent's own replies keep a row for each thing it did, such as Changed the plan or Told the subagents. Rehearsal writes the words of every row, never the model, so a row says only what happened, and the steps in a subagent's place are worded the same way. A row that acted on something names it as the page names it, such as Clicked Save changes. A row that did not work says so, such as Clicking Save changes did not work, and Re:agent reads the page as it is and carries on.

When a reply ends, its rows fold into one line, such as Worked for 1m 20s · opened a browser and took 14 steps. Press the line to open the rows again.

Stop or pause Link to Stop or pause

While the exploration runs, a strip above the message box says Exploring with the app's name and the time so far, and the stop button of the message box stops the exploration. Its label is Stop exploring. While Re:agent is writing a reply, such as an answer about the plan, the same button stops only that reply and leaves the exploration running.

When you stop, the subagents at work stop at a safe point, and everything they recorded stays with the exploration: the test cases, the bugs and the recordings. Re:agent then says the exploration stopped, and its report follows as it would at the end, as long as something was checked.

Pausing happens outside the app. rehearsal explore pause pauses an exploration at a safe point, and rehearsal explore resume carries on with the same brief, as the pause_exploration and resume_exploration tools do for a coding agent. The panel says Paused in the meantime, and the time spent paused does not count towards the exploration's minutes.

Good to know

The exploration runs on Rehearsal's servers, not in your tab, so you can close the tab and come back. The chat reads back everything that happened, and when the exploration ends on its own, Rehearsal tells you under the bell, and by email if you are away.

Next steps Link to Next steps