Agent Detail Dashboard
Use the Agent Detail Dashboard to look closely at a single agent. You can filter its recent runs, see why runs failed or were blocked, act on alerts, and approve or dismiss runs that were flagged for review. It's a good check before you rely on a new version of an agent.
To open the screen, go to Agent Ops → Agent Detail Dashboard.
Choose an agent
- In Agent, select the agent. You can type in the list to search for an agent by name.
- Review the dashboard. The first run in the list opens in Trace Detail automatically.
Only active agents and workflows appear in the list. Model shows the model the agent uses. You can't change it here.
Read the readiness score
The badge at the top shows a readiness score as a percentage. It's the share of runs in view that finished without a problem. The label beside it tells you where the agent stands:
- ready when the score is 90% or higher.
- needs review when the score is from 70% to 89%.
- action required when the score is lower.
- waiting for runs when there are no runs to score.
Point at the badge to see what is affecting the score, such as failed runs, prompt safety, tool governance, rate limits, runs waiting for review or alerts, with a suggested action for each.
Filter the runs
The filters apply straight away and update every panel on the screen.
| Filter | Options |
|---|---|
| Window | How many recent runs to load: 50 runs, 100 runs or 200 runs. |
| Status | All, Success, Failed or Blocked. |
| Risk | All, Prompt risk, Tool block, Rate limited or Alert. |
| Time | All, Last 1h, Last 24h or Last 7d. |
| Failed stage | Where a failed run stopped, such as Request validation or Response generation. |
| Error Type | All, Validation, Authorization, Timeout, Upstream, Persistence or Runtime. |
| Retryable | All, Retryable or Non-retryable. |
| Search | Find runs by conversation, user or error text. |
To go straight to one run, type its ID in Conversation ID and select Open Run. Select Refresh to reload everything.
Run Info
The Run Info tab shows totals for the runs in view: Runs, Errors, Avg latency, Cost, Input tokens, Output tokens, Prompt risks, Tool blocks, Rate limited and Alerts. Cost is an estimate worked out from the run tokens.
Alerts
The Alerts tab lists Triggered Alerts for the agent that are open or acknowledged. Each alert shows its message, severity, status, the rule that triggered it and the actual value.
- Select Ack to acknowledge an alert.
- Select Resolve when the problem is fixed.
- Select Dismiss if the alert doesn't need action.
Alerts raised by a single run are listed as run alert. Select one to open that run's trace.
If no alerts have fired, you see No triggered alert events for this agent yet. You set up alert rules when you edit the agent.
Quality
The Quality tab shows the Pending Eval Queue: runs that were flagged for a person to check, such as a blocked prompt, a failed tool call, a failed output check or user feedback. Each item says why it was flagged and what to check.
- Select Open Trace to see what happened in the run.
- Choose a decision:
- Approve if the output is fine.
- Corrected if you fixed the problem.
- Escalate if someone else needs to look at it.
- Dismiss if no action is needed.
The item moves to Reviewed, which lists past decisions. If nothing needs checking, you see No runs in view need review.
Governance
The Governance tab shows Tool Controls: the tools the agent used, with each tool's status and how many calls and failures it had. This list is read-only.
Error Breakdown groups failures by where they happened and their type, with how many there were and how many can be retried.
Charts
- Latency shows the response time of the latest runs. Point at a bar to see the run and its latency.
- Safety Mix counts prompt, tool, rate-limit and alert signals. Under it, Recent Errors and Alert Events list the latest failures and alerts.
Recent Runs
The Recent Runs table lists the runs that match your filters.
| Column | What it shows |
|---|---|
| Run | The run title or ID, with the agent name. |
| Status | Whether the run succeeded, failed, was blocked or is still pending. |
| Latency | How long the run took. |
| Cost | The estimated cost of the run. |
| Tokens | Tokens used. |
| Safety | Badges such as Prompt risk, Rate limited or Clear. |
| Updated | When the run last changed, in your own time zone. |
Select a row to open the run in Trace Detail. See Inspect a run.
Troubleshooting
| Message | What to do |
|---|---|
| Choose an agent to see its runs. | Select an agent in Agent. |
| No runs match the current filters. | Set some filters back to All, or choose a wider Time or Window. |
| No runs recorded for this agent yet. | Ask the agent a question in Agent Chat, then select Refresh. |
| Runs could not be loaded. | Select Refresh. |
| Alert events could not be loaded, so only run alerts are listed. | Select Refresh after a moment. |
If the problem continues, contact support.