Test signals on the agent page
While you edit an agent, you want to know whether its tests still vouch for it - without leaving the page to find out. The agent's page tells you in two places: the Tests button at the top, and a banner when the agent's latest run had failures.
Both are about the agent you have open, not the whole bot. A run can cover several agents; you only see this agent's share of it. For what to test and how to read a run, see Get started.
The Tests button
The Tests button shows how many tests belong to this agent. A dot on it gives you the state at a glance - green when the latest run passed, red when it had failures, amber when the results are outdated - and it pulses while a run is under way. Click it to see more:
| What you see | What it means | What to do |
|---|---|---|
| Last run with passed and failed counts | How this agent's tests did in the latest run. | View report to read the failures. |
| Run in progress | A run that includes this agent's tests is under way. | View progress to follow it. |
| Outdated - "Last run 3 days ago, before your last changes" | You've edited the agent since the latest run, so the results describe the agent as it was. | Re-run tests before you rely on them. |
| Drafted, never run | Tests are linked to this agent but have never run - common right after the Nexus AI drafts them. They don't prove anything yet. | Review tests, then run them. |
| No recent run | This agent has tests, but none of the bot's recent runs included them. They may have run earlier. | Open the Testing Lab to check. |
| Generating tests… | The Nexus AI is still drafting tests for this agent. | Review them once they're ready. |
| No test cases linked to this agent yet | Nothing tests this agent. | Create test case starts a new test with this agent already chosen. |
Run Tests (or Re-run tests) takes you to the Testing Lab with this agent's tests selected; the run starts when you confirm it there. View test cases opens the Testing Lab showing only this agent's tests.
Sub-agents don't have their own Tests button.
What Outdated means
Outdated means the agent changed after the run, not that the run went wrong. Results always describe the agent as it was when the run happened. A run doesn't record which version of the instructions it used, so Nexus never claims a run reflects the instructions you have now - only whether it came before or after your last change.
Treat outdated results as unknown. A test that passed before your edit may not pass after it, so run the tests again before you act on them.
The failing-tests banner
When this agent's latest run had failures, a banner at the top of the agent's page says how many - for example 3 of 12 tests failing - and when the run happened. View report opens the run.
- It's about this agent only. An agent whose own tests passed doesn't show it, even when other agents in the same run failed.
- It counts only real failures. Tests the run couldn't finish (Systematic error) say nothing about the agent, so they aren't counted. A run where every failure was one of those shows no banner.
- It tells you when the results are outdated. If you've edited the agent since the run, it says the run was before your last changes - read those failures as describing the older agent.
- Closing it lasts for that run only. Once you've seen it, close it; the next run with failures shows it again, so one dismissal never hides a later regression.
To clear it, fix the agent or the tests, then run them again. Debug a failing test case walks through finding what went wrong.
Next: Debug a failing test case - find the step that went wrong and fix it.