When you turn on Long Running Tasks in Agent Builder, an eligibility check evaluates the saved system prompt and the selected tools. If the agent does not qualify, the switch returns to off, and the Long Running Tasks panel shows a status tag and a message that explains the reason. If the check cannot finish, the panel shows the Check failed status tag and a Validate again button.

For the conditions that an agent must meet, see Enable long-running tasks for an agent.

Agent is not a record-processing task

Error: This doesn’t look like a record-processing task

What this error means

The panel shows the Not eligible for LRT status tag and a red message with this heading. The message describes what the check found in the system prompt, such as a read-only conversational assistant that presents data to users. It then explains that long-running tasks suit agents that change or resolve records, not agents that write an answer for someone to read, and asks you to describe what each record should end up with. The switch returns to off, and the capability row shows Not eligible.

Why the error occurs

This occurs when the agent produces an answer for a person to read, does not operate on records, or writes only to configuration.

How to fix the issue

To resolve the issue, update the system prompt and run the check again:

• Go back to the step where you enter the system prompt.

• State which records the agent processes and what each record should end up with, such as the agent creating, updating, merging, classifying, relating, or resolving the record.

• Save the draft, return to the Capabilities step, and turn on Long Running Tasks again.

Preventive measures

To avoid this issue, follow these recommendations:

• Describe the agent's work as an operation on records that produces a data outcome.

• For more information about writing the system prompt, see AgentFlow system prompt guidelines.

Every record requires a human decision

Error: REQUIRES_PER_RECORD_HUMAN_DECISION

What this error means

The panel shows the Not eligible for LRT status tag and a red message that explains the reason. The switch returns to off.

Why the error occurs

This occurs when every record needs human judgement, the system prompt states no rule for making that decision, and the system prompt does not describe an automated mode.

How to fix the issue

To resolve the issue, update the system prompt and run the check again:

• State the rule that the agent uses to decide what to do with each record.

• If the agent asks for confirmation before it acts, describe an unattended or automated mode in which the agent proceeds without confirmation.

• Save the draft, return to the Capabilities step, and turn on Long Running Tasks again.

Preventive measures

To avoid this issue, follow these recommendations:

• If the agent asks for confirmation, describe in the system prompt how the agent proceeds when it runs unattended.

Instructions do not say what to do with each record

Error: NO_PER_RECORD_ACTION

What this error means

The panel shows the Not eligible for LRT status tag and a red message that says what is missing. The switch returns to off.

Why the error occurs

This occurs when the system prompt shows that the agent operates on records, but does not say which steps the agent applies to one record.

How to fix the issue

To resolve the issue, update the system prompt and run the check again:

• List the steps that the agent applies to one record, in order.

• State that the agent repeats the same steps for each record.

• Save the draft, return to the Capabilities step, and turn on Long Running Tasks again.

Preventive measures

To avoid this issue, follow these recommendations:

• Write the per-record steps as a numbered workflow in the system prompt. For more information, see AgentFlow system prompt guidelines.

Selected tools cannot do the work

Error: MISSING_REQUIRED_TOOLS

What this error means

The panel shows the Not eligible for LRT status tag and a red message with this heading. The message tells you that your instructions describe work that the current tool selection cannot carry out. The switch returns to off.

Why the error occurs

This occurs when the system prompt describes work on records, but no selected tool can perform that work. For example, no tools are selected.

How to fix the issue

To resolve the issue, update the tool selection and run the check again:

• Go to the Tools step.

• Select the tools that can read and act on the records that the task needs.

• Save the draft, return to the Capabilities step, and turn on Long Running Tasks again.

Preventive measures

To avoid this issue, follow these recommendations:

• Select the tools before you turn on Long Running Tasks.

• Run the check again after you add or remove tools, because the check result applies to the tools that were selected when the check ran.

Eligibility check could not complete

Error: EVALUATION_SERVICE_UNAVAILABLE (Transport error, or a reply that will not parse)

Error: EVALUATION_TIMED_OUT (The 150-second server deadline elapsed)

Error: TOOL_CATALOG_UNAVAILABLE (Tool descriptions could not be loaded)

What this error means

The panel shows the Check failed status tag, a red message with this heading, and a Validate again button. The message adds a reason and asks you to try again in a moment. For example: “We couldn’t load your tool list to check against. Try again in a moment.” The switch returns to off.

Why the error occurs

This occurs when the check service is unavailable, the tool list cannot be loaded, or the check does not finish within 3 minutes.

How to fix the issue

To resolve the issue, run the check again:

• Wait a moment, and then select Validate again.

• If the check continues to fail, contact Reltio Support. Include your tenant ID, the agent ID, the exact message, the time of the failure, and the steps that you already tried.

Preventive measures

To avoid this issue, follow these recommendations:

• Select Validate again once, and wait for the result. Only one check runs at a time for an agent.

Instructions and tools selection changed

Error: Your instructions and tools selection changed

What this error means

The panel shows a red message with this heading, the text “Run validate again”, and a Validate again button. The switch stays on, but the last check result is out of date.

Why the error occurs

This occurs when the draft changed after the last check ran.

How to fix the issue

To resolve the issue, run the check again:

• Select Validate again, and wait for the check to finish.

• If the check does not pass, resolve the issue that the panel shows, and then run the check again.

Preventive measures

To avoid this issue, follow these recommendations:

• Finish editing the system prompt and the tool selection before you run the check.

• After you edit the system prompt or the tools on another step, return to the Capabilities step and run the check again.

Submit for review is blocked

Error: Long Running Tasks is on, but the draft changed after the last eligibility check. Run the check again before submitting.

What this error means

Agent Builder does not submit the agent for review. The agent stays in Draft state, and the security scan does not run.

Why the error occurs

This occurs when you submit an agent that has Long Running Tasks turned on and the check result is out of date. Agent Builder also blocks the submission when a check is still running, when no check has run, or when the check has failed.

How to fix the issue

To resolve the issue, run the check and submit the agent again:

• Go to the Capabilities step.

• Select Validate again, and wait for the status tag to read Eligible for LRT.

• Submit the agent for review again. For more information, see Create, test, and submit an agent for review.

Preventive measures

To avoid this issue, follow these recommendations:

• Run the check after your last edit and before you submit the agent for review.