Workbench
The Workbench is an Agent-driven collaboration space for DataOps tasks. Around a clear task goal, the Agent collects information, analyzes the issue, and compiles the results according to the selected execution mode. You can also put related tasks in the same workspace to follow up on and review them together.
The Workbench is currently in the trial stage. It first opens up the ops troubleshooting scenario: starting from a failed data task, it automatically collects evidence, locates the cause, and gives handling suggestions.
1. Understand the Workbench, workspaces, and troubleshooting tasks
The trial version currently supports troubleshooting two types of alerts: Flow run failed and Task run failed. Each failure corresponds to one troubleshooting task, which records the original alert, execution process, evidence, conclusion, and handling suggestions. Plan mode also generates a troubleshooting plan.
Three core objects
| Object | Description | How to use it |
|---|---|---|
| Abnormal event | Generated by failure alert rules; the starting point of troubleshooting | View the failed object, time of occurrence, and alert context |
| Troubleshooting task | Diagnosis of a single abnormal event | Depending on the execution mode, view progress, a report or troubleshooting plan, and the evidence, conclusion, and suggestions |
| Workspace | A container that organizes related troubleshooting tasks; not the same as a DataOps space | Summarize the progress of the same issue or business pipeline and generate a workspace briefing |
When creating a task, choose an execution mode based on whether you need to confirm a plan, review evidence step by step, and provide additional information interactively:
| Execution mode | Process after starting | Plan confirmation | Follow-up review and collaboration |
|---|---|---|---|
| Lightning mode | Runs directly without generating a plan, then compiles the task conclusion | No confirmation required | View analysis progress and the report; does not use the follow-up Q&A or step expansion process |
| Plan mode | Generates a plan first, then troubleshoots step by step after confirmation | Manual confirmation | View the troubleshooting graph, steps, and evidence, and provide additional information when prompted |
| Plan mode · Auto confirm | Generates a plan first, then runs after automatic confirmation | Automatic confirmation | Uses the troubleshooting and collaboration process of Plan mode |
Manually created tasks use Plan mode by default. Plan mode · Auto confirm still generates a plan and is different from Lightning mode. Existing tasks do not switch modes automatically when Lightning mode is introduced.
How to divide workspaces
Put related tasks for the same issue, the same business pipeline, or anything that needs continuous follow-up in the same workspace, so you can use the briefing to spot common causes and overall progress. Create separate workspaces for unrelated issues. After an issue is resolved, you can archive the workspace so that historical tasks don't interfere with current troubleshooting.
2. Start an ops troubleshooting task
Before you start
Make sure at least one failure alert rule is enabled in the space. The Workbench only shows anomalies found by alert rules. Without alert rules, there are no abnormal events to troubleshoot.
Select an anomaly and a workspace
Open the Workbench, click New task, select the anomaly to troubleshoot, and decide whether to put it in an existing workspace or create a new workspace. If the same issue already has a troubleshooting task, add it to the original workspace first. If the issue is independent, create a new workspace.
View details. Before creating the task, check the failed object, alert time, and related information to avoid troubleshooting the wrong anomaly.
Add diagnostic context. If you've recently changed code, dependencies, or configuration, or have a clear expectation of the normal result, describe it in the additional notes before creating the task. All three execution modes use this information.
Choose an execution mode. In the execution mode dropdown in the creation area, select Lightning mode, Plan mode, or Plan mode · Auto confirm. Choose Lightning mode to view analysis results directly. Choose Plan mode if you need to review the plan first or provide information through Q&A during the process. After checking the information, click Create.
Start execution and confirm the plan
After you create a task with Lightning mode selected, the Workbench starts execution directly without waiting for a troubleshooting plan. Open the task to view analysis progress and results.
With Plan mode, first review the generated troubleshooting plan, focusing on the goal, scope, and known context. If anything is missing, add background or adjust the goal before confirming execution. With Plan mode · Auto confirm, the plan is confirmed automatically after it is generated, and execution starts.
3. View the troubleshooting process and analysis results
Review the process for the selected execution mode
Lightning mode: view progress and the report. While the task runs, the page shows analysis progress. After the results are generated, you can view the analysis report and expand "N thinking steps" to review the progress so far. This count represents the progress records shown, not the number of nodes in the Plan mode troubleshooting graph.
Lightning mode skips plan generation, plan confirmation, and step-by-step review, but still compiles a task-level conclusion after execution. If the report content is visible while the task status is still "Generating conclusion", wait for the conclusion to finish and assess the result together with the final task status.
Plan mode: view the troubleshooting graph and step evidence. In Plan mode and its automatic confirmation option, the Agent follows this process:
- Generates a troubleshooting plan based on the abnormal event and the additional information.
- Queries task instances, run logs, metrics, change records, and other information according to the plan.
- Reviews the existing evidence after each step. When clues are insufficient, adds troubleshooting steps or requests information from the relevant people.
- When the evidence is sufficient, compiles the failure cause, scope of impact, and handling suggestions.
Use the troubleshooting graph to view diagnostic nodes and steps. Select a node to view the query purpose, execution result, and evidence source, so you can judge how the conclusion was reached.
How to collaborate on troubleshooting
Add information based on evidence. When Plan mode shows a "Needs more information" reminder, provide the business background, expected result, recent changes, or known scope of impact. The troubleshooting continues based on the new information. After a task in Lightning mode is created, it does not support continuing troubleshooting through follow-up Q&A, expanding steps, or changing the execution mode within the same task. Add all known context before creating the task.
Go back to the original object to check. When you need to confirm the full logs or task configuration, use the related links in the task to open the task instance or asset.
Handle loading failures and stop troubleshooting. If a Lightning mode report fails to load, click Retry to load the report again. This does not rerun the troubleshooting. To stop a running task, click Terminate and confirm. After termination, you can review the progress so far, but a new report is not guaranteed. You can archive a task as needed after it completes, fails, or is terminated.
Handling boundaries
The Workbench diagnoses issues and gives suggestions, but does not run fixes automatically. "Direct execution" in Lightning mode means running the analysis directly, not automatically fixing the failure. Operations such as rerunning tasks, modifying code, adjusting dependencies, or changing configuration still need to be confirmed and completed by people with the required permissions.
4. Use workspace briefings to track overall progress
The workspace briefing aggregates information from the unarchived tasks in the current workspace. It helps you quickly tell whether an issue already has a conclusion, what needs to be handled next, and whether multiple troubleshooting tasks point to a common cause.
| Area | Question it answers | Suggested action |
|---|---|---|
| Identified | Which issues already have a clear cause or conclusion | Check the conclusion and confirm whether handling is complete |
| Handling suggestions | Which follow-up actions can be taken now | Assign an owner and open the corresponding task to view the basis |
| Key points | Which clues, symptoms, or patterns are worth watching | Judge whether multiple anomalies share a common cause |
After tasks are added, change status, or finish troubleshooting, you can update the briefing to get the latest summary. The briefing only counts unarchived tasks. "Handling suggestions" focuses on suggestions from recently created tasks.
Go from the briefing back to tasks
The briefing is an overview and doesn't replace the full evidence of individual tasks. When you need to confirm the basis of a suggestion, use the related task count or the corresponding entry to open the specific troubleshooting task and view its execution process, evidence, and context.
5. Configure automatic creation and execution modes
When to turn it on
For production environments that need continuous operations and should go into diagnosis right after a failure, we recommend turning on automatic creation. It reduces the time spent waiting for someone to notice anomalies and create tasks repeatedly.
Automatic creation determines when troubleshooting tasks are created, while the execution mode determines how they investigate. Turning on automatic creation does not select Lightning mode.
| Setting | Purpose | Configuration focus |
|---|---|---|
| Automatically create tasks | Automatically creates a troubleshooting task after a new failure alert occurs | After enabling it, select the execution mode and workspace policy |
| Execution mode | Select Lightning mode, Plan mode, or Plan mode · Auto confirm | Lightning mode runs directly; Plan mode waits for confirmation; Plan mode · Auto confirm still generates a plan first |
| Run in | The system creates workspaces automatically, or tasks always join a specified workspace | Specify a fixed workspace when you're continuously focusing on the same business |
Open the Automatically create tasks settings and turn on the switch. Select the mode you need under Execution mode, configure Run in, and click Save. Lightning mode does not require plan confirmation. After changing the options, click Save to submit the settings. Clicking Cancel does not save them. Changing the execution mode setting does not switch existing tasks to another mode.
Effective rules
After it's turned on, only new failure alerts generated afterward that haven't been handled yet are processed. No tasks are backfilled for historical alerts. Automatically created tasks are assigned to the asset owner first. If the specified workspace has been archived, it's restored and used. If it has been deleted, the system falls back to system-created workspaces.
We recommend turning on both automatic creation and external notifications: the system continuously finds and diagnoses issues, and the relevant people receive timely reminders when Plan mode needs additional information, handling suggestions are ready, or results for a task in Lightning mode are ready.
6. Handle tasks promptly through @Me and external notifications
@Me collects the troubleshooting reminders you need to follow up on. It distinguishes pending and read status, and you can open the corresponding task directly.
| Reminder type | When it appears | What to do |
|---|---|---|
| Needs more information | The existing evidence in Plan mode is insufficient and more business context is needed | Add the expected result, recent changes, or scope of impact |
| Handling suggestions ready | Troubleshooting already has actionable suggestions | Review the basis of the suggestions and arrange follow-up handling |
| Lightning mode results ready | A task in Lightning mode has finished compiling its results | View the analysis report and check the evidence and conclusion |
The results-ready reminder for tasks in Lightning mode asks you to view the report. It does not mean that you need to answer a question or that the failure has been fixed. For manually created tasks, the task creator is reminded by default when handling suggestions or Lightning mode results are ready. For automatically created tasks, the asset owner is reminded first.
Get reminders even when you're away from the Workbench
To receive messages through external channels such as Feishu, a space admin needs to enable the Workbench task diagnosis alerts notification rule, allow personal subscriptions, and bind a notification channel to the DataOps space. Users then turn on personal subscriptions as needed.
If you use custom notification conditions, ask an administrator to confirm that they include results-ready reminders for tasks in Lightning mode. Conditions that cover only requests for more information or handling suggestions cannot receive these result notifications.
If you don't receive external reminders
- Confirm that the Workbench task diagnosis alerts rule is enabled in the space.
- Confirm that the admin allows personal subscriptions and that you've turned on your subscription.
- Confirm that an available notification channel is bound to the current DataOps space.
- If you do not receive a result notification for a task in Lightning mode, confirm that the custom notification conditions include results-ready reminders.
If you don't have management permissions, contact a space admin to complete the configuration above.

