Debug and run with Copilot
Run a test case’s automated steps live, watch each one in the browser, and fix a step where it fails. Every Copilot session uses the same window, whether it started from learning, recording, or debugging.
The agents behind a run
Section titled “The agents behind a run”Three agents step in when a step fails. They work in Copilot sessions and in test plan runs.
- Healer: finds the element another way when every saved locator stops matching, and keeps the step running. You decide whether to keep the healed locator.
- Analyzer: explains why a step failed: a verdict, the root cause, and what to change.
- Bug Reporter: raises a failed run as a defect in your issue tracker, written up from the run.
Start a session
Section titled “Start a session”| Where | Button | What it does |
|---|---|---|
| A test case that has automated steps | Debug with Copilot | Opens the test case to run and edit its steps |
| A run result | Run with Copilot | Replays every step live, without learning |
| A run result’s more options icon (⋯) | Debug with Copilot | Opens a session on that run |
In the Copilot configuration, choose Where to run. To stop on a step, set Pause at under Debugging. Then click Start Copilot. See Copilot configuration.
The session opens over the test case or in a new tab. Debug with Copilot runs the steps as soon as the session starts, and pauses at the end of the test, or at the step you set in Pause at.
The session window
Section titled “The session window”- Steps panel: the test case’s steps, with the toolbar above them. Click the Chat icon at the top of the panel to open Chat below the steps, with Copilot’s messages and a prompt box for instructions. Type
/for commands. - App: the app under test, labeled Lab machine on Testsigma Lab or Live browser on your own machine. Click Inspect to look at an element. The header shows the session’s timer.
- Run log: a drawer with everything the run reports. See Use the Run log.
Arrange the steps panel
Section titled “Arrange the steps panel”Click the layout icon next to the session’s title and select how the steps panel sits:
- Docked to the side: the steps sit beside the app. This is the default.
- Floating overlay: the steps float over the app, so the app gets the full width.
- Minimize to a pill: the steps shrink to a small card with the session’s state, the current step, such as Step 13 of 13, and Resume. Click the expand icon on the card to bring the steps back.
Use the Run log
Section titled “Use the Run log”The Run log drawer lists each step as it runs, with the time, the step, its result, such as passed, and how long it took. Click the bar to open it.
- All and Errors {n}: show every entry, or only the errors.
- Log file: downloads the run’s log as a text file. It shows under All.
- Dock side: click the dock icon and select Dock left, Dock bottom, or Dock right. Press
⌘+Shift+Don macOS orCtrl+Shift+Don Windows to move the drawer to the next side. - Resize: drag the drawer’s edge.
- Collapse: click the chevron icon (∨) at the top of the drawer.
Arcus keeps the drawer’s side, size, and whether it’s open for your next session in this browser. Web and mobile sessions keep their own layout.
In a mobile session, the drawer is open by default and has more tabs:
- Element: the screen’s element tree and the attributes of the element you select.
- Run log: the same log as in a web session.
- Device log and Network: not available yet. They show “Device logs will come in the next release.” and “Network logs will come in the next release.”
- Source code: the source of the current screen.
Run, pause, and step
Section titled “Run, pause, and step”| Button | What it does |
|---|---|
| Resume | Continues the run from the paused step |
| Retry | Shows in place of Resume after a step fails. Runs the failed step again and continues |
| Pause | Stops the run at the current step |
| Step over | Runs the next step, then pauses |
| Skip over | Skips the next step without running it |
| Restart execution | Runs every step again from the start |
| Record | Records what you do in the browser as steps. See Record and author. |
The status bar at the bottom of the steps shows where the run is:
- Execution stopped. Step n failed.: a step failed. Fix it, then click Resume or Restart execution.
- Paused at debug point. Step n: the run stopped at the step you set in Pause at.
- Paused at the end of the test: every step ran.
Point at a step to show its more options icon (⋮). The menu has Run from here, View results, Step settings, Move up, Move down, Duplicate, Skip this step, Insert a step above, Insert a step below, and Delete.
To see what a step did, click View results. Step results shows the step’s result, any heal card, the screenshot from this run with Compare steps, and the element and locator the step used.
When a step fails or heals
Section titled “When a step fails or heals”When a step fails, in a web or a mobile session, the drawer opens on Run log > Errors.
When the Healer runs on a step, the step list marks the step while it looks for the element, and again when the step heals or the heal fails. The chat shows the same progress, for example “Healer Agent is looking for the element on step 4…”, then “Healer Agent found the element on step 4” or “Healer Agent could not find the element on step 4”.
When the step finishes, Copilot adds cards to the chat:
- Healer card: the previous locator and the updated one, with actions to save the healed locator or dismiss the card. See Review a heal in a Copilot session.
- Analyzer Agent: step failed or Analyzer Agent: recovered without stopping: the Cause, the Analyzer’s Verdict, and the Action taken. See Where verdicts appear.
Add or fix a step while paused
Section titled “Add or fix a step while paused”- While the run is paused, click Type a step, or / for blocks below the last step, or click a step to change it.
- Type the step. As you type, Copilot suggests matching steps, such as Verify page or Verify element. Choose one.
- Fill in the parts of the step. For a value, choose its source: Plain text, Runtime, Parameter, or Environment.
- Click Resume to run the new step.
Until a step is complete, a banner shows 1 line is not saved: fix it to save it.
To point a step at an element on the page, use Pick from page. See Write automated steps.
End the session
Section titled “End the session”- Click the exit icon at the top of the steps.
- In End session and close this tab?, click End session.
Ending the session stops the run, and closes the browser on your machine or releases the cloud machine on Testsigma Lab. Steps that already ran keep their results, and your changes are saved to the test case. A session in its own tab closes the tab. The run appears in the test case’s Run history. See Run results.
Healer
Section titled “Healer”The Healer (auto-heal) keeps a step running when its element’s saved locators stop matching, such as after a button is renamed. It finds the element another way, confirms it against two other locators, and continues the run. You then accept the healed locator, or leave it.
How the Healer finds an element
Section titled “How the Healer finds an element”When a step runs, Arcus looks for its element in this order:
- The element’s saved locators, starting with the primary.
- A healed locator saved by an earlier run, if there is one.
- The element’s backup locators (the Backup Pool tab).
- Gen-AI, which looks for the element on the live page.
If a locator finds exactly one element, the step runs. When Gen-AI finds it, the Healer confirms the match with two other locators that must point at the same element (shown as Anchor 1 and Anchor 2), runs the step, and saves the result:
- Healed locator saved for this element: the next run tries it before the backup pool.
- Backup locators: a set of alternatives, each with its Source (GenAI auto-heal) and a Quality of Good, Weak, Fragile, or Not Validated.
The step passes. The primary stays as it was until you accept the healed locator (see Accept a healed locator).

The Healer doesn’t run when a saved locator matches more than one element. The step fails instead (see Read a step that didn’t heal).
Where the Healer runs
Section titled “Where the Healer runs”| Where | What controls it |
|---|---|
| Test plan runs | Auto-heal this element on the element. There’s no auto-heal setting on the run configuration, the run template, or the plan’s Automation Settings |
| Copilot sessions | Auto-heal this element on the element. The Copilot configuration has no self-healing setting |
The Healer doesn’t apply to Windows elements. On Windows, a locator that stops matching fails the step. See Elements and locators.
Turn auto-heal on or off for an element
Section titled “Turn auto-heal on or off for an element”-
On the Automated steps tab, click the element in a step, such as
@checkout/email. Step settings lists the element’s locators under Tried in this order on Web. -
Click Element details on Web. The Element details panel opens on the Locators tab.
-
Select or clear Auto-heal this element.

Auto-heal this element is selected by default for every element. When it’s selected and a locator stops matching, Arcus looks for the element on the live page rather than failing the step.
Element details opened from a run result or from a Copilot session is read-only. To change an element, open it from the test case.
Review a heal in a Copilot session
Section titled “Review a heal in a Copilot session”When a step’s locator stops matching in a Copilot session and the element has Auto-heal this element selected, the step list and the chat show the Healer’s progress (see When a step fails or heals). A step that heals takes longer than usual, because the Healer searches the page before the step runs.
-
When the step finishes, read the Healer Agent’s card in the chat. It shows how the element was found, the previous locator, and the updated one.
-
Save the updated locator from the card. The card then shows “locator saved on step n”.
Saving makes the healed locator the element’s primary, like Approve as primary on a run result (see Accept a healed locator). The change applies to every test case that uses the element. To hide the card without saving, dismiss it. Dismissing doesn’t reject the heal.
The card doesn’t offer to save when you can’t edit the test case, or when the run didn’t report the healed locator.
When a heal fails, the Analyzer card explains why. See Analyzer.
Review a healed step
Section titled “Review a healed step”-
Open the step’s result. In a plan run, open the run, click the test case, and click the step. In a Copilot session, click the ellipsis icon (⋯) on the step and click View results. A test case result with a healed step shows {n} step auto-healed in its header.

-
Read the ANALYZER VERDICT card. It shows Fixed, approval required, how the element was found (Healed with Gen-AI), a summary such as “The primary locator stopped matching; Gen-AI read the page and found the element another way.”, and the REASON.
-
Under Locator change, read the Previous · recorded locator, struck through, and the Current · applied locator. Confirmed by 2 other locators means two other locators point at the same element. Affects {n} tests shows how many test cases use the element.
-
(Optional) Click Locators tried to see every locator the step tried, or Compare to see the screen At the time of recording next to At the time of execution. The Element source tab shows the page source at execution and has Locate element.

Check a healed element
Section titled “Check a healed element”Until someone accepts the healed locator, Element details shows its state on the Locators tab:
- The primary shows Not matching since a date, with its Consecutive failures and Total failures.
- WORKING NOW: “Backup running in place of the primary”, with its type, source, quality, and anchors. Below it, a line reports progress, for example “Passed 2 runs in a row. Auto-replace is off, so it becomes the primary only when someone accepts it.”
- GEN-AI: “Healed locator saved for this element” and “The next run tries this before the backup pool.”
On a later run where the healed locator works, the step result shows Passed · no heal ran, even though the primary still doesn’t match.
Accept a healed locator
Section titled “Accept a healed locator”Accept a healed locator from a run result. In a Copilot session, save it from the Healer Agent’s card instead (see Review a heal in a Copilot session).
-
Open the run result and click the healed step.
-
On the ANALYZER VERDICT card, click Approve as primary. The change applies to every test case that uses the element.
To promote healed locators without anyone accepting them, select Auto-replace primary in Element details. A backup is then promoted once it passes 3 runs in a row.
If you don’t approve it, the primary stays as it is, and later runs keep trying the healed locator first.
Manage backup locators
Section titled “Manage backup locators”The recorder saves backups when it captures an element, listed with a Source of Smart Recorder, Classic Recorder, or Semantics. The Healer adds more. Open the Backup Pool tab in Element details to see every backup, with its Type, Source, Quality, Last Check (Not Checked, Matched, or Not Found), and Results (passes out of runs, such as 1 / 1). Filter with All, Matching, Not matching, Unchecked, or Retired, or click Group by Source. The tab shows when it was last updated.
Quality grades how likely a backup is to keep finding its element. Arcus tries backups with a better grade first, in the order Good, Weak, Not Validated, Fragile. When an element has more backups than it can keep, the ones with the worst grades are dropped first.
To add a locator manually, select a type under Add a locator for Web (CSS Selector, XPath, Name, Class, Tag, Link text, or Partial link text), enter the value, and click Add. ALL LOCATORS, IN ORDER lists every locator in the order Arcus tries them.
History in Element details shows COMING SOON. Changes to an element, including heals, aren’t listed there yet.
Read a step that didn’t heal
Section titled “Read a step that didn’t heal”The Healer couldn’t find the element
Section titled “The Healer couldn’t find the element”When the element isn’t on the page at all, every locator and Gen-AI miss, and the step fails with NO_SUCH_ELEMENT. The ERROR block says how many alternatives the Healer tried, for example “Self-healing tried 2 alternatives without a match.” Below it, the ANALYZER VERDICT explains why, such as a test that reached the wrong page, with the REASON, what to change, and Debug with Copilot. See Analyzer.
On the Backup Pool tab, every backup’s Last Check changes to Not Found.
This usually means the step ran on the wrong page or the app changed more than a locator. Click Debug with Copilot to replay the run up to the step and see what’s on screen.
A locator matched more than one element
Section titled “A locator matched more than one element”A step fails without healing when a saved locator matches more than one element. The step result shows the card The primary locator missed, “The primary locator found nothing on the page and no heal ran for this step, so it failed.”, followed by the error and every locator tried, with Open element.
The error reads, for example, “5 visible elements match this element, and the step needs exactly one. Re-record the element against the one you mean.” The error code is MULTIPLE_ELEMENT. Re-record the element so its locators point at one element, or turn on Pause on ambiguity in the Copilot configuration so that Copilot stops and asks.
A step whose locators matched normally shows Passed · no heal ran: “Auto-heal did not run for this step, so no backup locator was tried and nothing was healed.”

Analyzer
Section titled “Analyzer”The Analyzer explains why a step failed. It reads what the run captured for the step, such as the screenshot, the locators tried, and the earlier steps, and gives a verdict with the root cause and what to fix. The error on the step tells you what the engine reported. The Analyzer verdict tells you why.
What a verdict contains
Section titled “What a verdict contains”A verdict appears below the step’s ERROR block, under ANALYZER VERDICT:
- A chip that says what to do next, such as No automatic fix this release. See Verdict chips.
- A one-line title that names the failure, for example “Step 3 click failed: landed on example.com placeholder instead of arcus app”.
- A summary of what went wrong.
- REASON: the root cause, from the evidence.
- What to change to fix the test.
- The evidence the verdict is based on, such as which earlier steps passed and what the screenshot shows.
- Debug with Copilot, which opens a session on the run so you can see the step and fix it.

While the Analyzer reads the evidence, the block shows “Reading the evidence for this step. This usually takes under a minute.”
Proposed fix
Section titled “Proposed fix”When the Analyzer can write the fix itself, the verdict ends with a PROPOSED FIX: the step as it ran under Previous · as run, and the step as it would be under Proposed, in Arcus step syntax. Confidence gives how sure the Analyzer is, as a percentage.
Arcus never applies a proposed fix on its own. Under Why it waited for you, the card says that fixes suggested while explaining a failure always need your review. Click Apply, then Apply again to confirm, and Arcus changes the step in the test case for every run from now on. If the step is in a step group, the fix changes the group and every test that calls it. The card then shows Applied. Click Dismiss to leave the step as it is. You can give a reason.
The verdict’s suggested actions are written in Arcus step syntax too, so you can paste one into the step. Ask the agent in the panel’s footer isn’t available yet; it reads “Chatting with the analyzer will be available soon.”
Where verdicts appear
Section titled “Where verdicts appear”- Run results: the ANALYZER VERDICT block on a failed step. See Run results.
- Copilot sessions: a card in the chat when the Healer asked the Analyzer about a step. It’s titled Analyzer Agent: step failed, or Analyzer Agent: recovered without stopping when the heal let the step pass. It shows the step number, the Cause, the Verdict, and the Action taken.
Verdict chips
Section titled “Verdict chips”| Chip | What it means |
|---|---|
| Report this as a defect | The app is wrong, and the step’s expectation holds. See Bug Reporter. |
| Fix options — you decide | The Analyzer has fixes for you to choose from, often with a proposed fix. |
| Expected result changed | The app’s output no longer matches the step’s expected value, such as a page title. Review the expected value; a proposed fix usually updates it. |
| Relearn element | The element wasn’t found, or its locator is no longer valid. Re-record the element. |
| Refine the assertion | The step checks the wrong element or row. Change what the step verifies. |
| No automatic fix this release | The app’s flow changed before this step. Update the steps manually. |
| Recommendation | The Analyzer suggests a fix. |
| Undetermined | The Analyzer couldn’t reach a conclusion. |
When the Healer asked the Analyzer about the step, a second chip names the failure, such as Locator changed, Element not found, Flow changed, Page still loading, or Application defect.
Which steps get a verdict
Section titled “Which steps get a verdict”The Analyzer explains a failed step, once the run has finished. It doesn’t explain:
- A step group, a condition, or a loop. It explains the step inside it that failed.
- An API call, such as a REST, GraphQL, or SOQL step.
- A step that calls a step group.
- A step that passed.
- Any step while the run is still queued, running, or paused.
Arcus can update a verdict after the run, as the Analyzer revises its analysis. A run result shows the latest verdict.
The Analyzer and the Healer
Section titled “The Analyzer and the Healer”When a locator stops matching, the Healer looks for the element another way. Before a healed locator is applied, the Analyzer checks whether the failure is a locator problem. If it isn’t, such as a step that ran on the wrong page, the heal isn’t applied and the step fails. If the Analyzer doesn’t answer in time, the heal is applied.
In a Copilot session, the Analyzer card shows when this happened. Its Verdict ends with “Not a locator problem, so healing cannot fix it.”, and Action taken reads “Nothing was changed.” When the Analyzer didn’t answer in time, Verdict reads “No verdict: the analyzer did not answer in time.”
Bug Reporter
Section titled “Bug Reporter”The Bug Reporter raises a failed run as a defect in your issue tracker. It writes the title from the step that failed, lists the steps up to the failure and the engine’s error in the description, and attaches the failed step’s screenshot. You check the write-up and create the issue, or link one that already exists.
It starts from a run result, not from the Copilot session: click the defect icon on the test case’s row in the run, or click Raise it as a defect in the result’s more options icon (⋯), under SHARE. When a verdict in the session reads Report this as a defect, open the run result and report it from there. See Report a defect from a run.
Troubleshooting
Section titled “Troubleshooting”The browser exits before the session starts
Section titled “The browser exits before the session starts”The message says the browser exited before it opened its debug port. Choose Chrome instead of Chrome for Testing under Browser on this machine, and start the session again.
The browser was closed
Section titled “The browser was closed”If the browser closes during a session, click Open browser to start it again.
Was this page helpful?
Thanks for the feedback.