Run Debugger

Introduction

The Run Debugger allows administrators to investigate assignment executions in detail. When something goes wrong—or when you need to understand exactly what an assignment did—the debugger provides the visibility you need.


Accessing the Run Debugger

  1. Navigate to the Admin section

  2. Select Run Debugger

  3. Enter the identifier for the job you want to investigate

You can search by:

  • Job ID - The unique identifier for a specific execution

  • Assignment ID - Find jobs for a specific assignment

  • Sandbox ID - Technical identifier for the execution environment


What You Can See

Execution Timeline

Follow the assignment's execution step by step:

  • When the job started

  • Each action the assignment took

  • When connections were called

  • How long each step took

  • When and how the job completed

Messages and Responses

See the conversation between the assignment and the AI:

  • SOP that was provided

  • How the AI interpreted the task

  • Decisions made during execution

  • Responses and outputs generated

Tool Calls

View every connection and tool interaction:

  • Which connections were used

  • What actions were taken

  • Parameters sent to each tool

  • Results returned from each call

Errors and Failures

When things go wrong, see exactly what happened:

  • Error messages and codes

  • Where in the execution failure occurred

  • Context around the failure

  • Information to help diagnose the cause


Debugging Common Issues

Assignment Stopped Unexpectedly

Check for:

  • Timeout issues (job took too long)

  • Error responses from connections

  • Unexpected data that caused confusion

  • Missing permissions or access

Wrong Output Produced

Review:

  • Input data the assignment received

  • How SOP was interpreted

  • Decisions made during execution

  • Where the logic went wrong

Connection Failures

Look at:

  • Authentication status

  • API responses from external services

  • Rate limiting or quota issues

  • Network or connectivity problems

Human-in-the-Loop Not Working

Verify:

  • Request was properly created

  • Notification was sent

  • Response was received and processed

  • Assignment resumed correctly


Using Debug Information

For Immediate Fixes

When you find an issue:

  1. Identify the root cause from logs

  2. Determine if it's configuration, connection, or SOP related

  3. Make the necessary fix

  4. Re-run to verify the solution

For Pattern Recognition

Review multiple jobs to spot:

  • Recurring issues

  • Edge cases not handled

  • Performance bottlenecks

  • Opportunities for improvement

For User Support

Use debug information to:

  • Explain what happened to users

  • Show them where their assignment needs adjustment

  • Guide them to better configurations


Best Practices

Start with Recent Jobs

When investigating issues, look at:

  • The specific failed job

  • Recent successful jobs for comparison

  • Any pattern in failures

Follow the Timeline

Walk through execution chronologically:

  • What was the assignment trying to do?

  • Where did things go as expected?

  • Where did problems start?

  • What was the final outcome?

Note Key Findings

Document what you discover:

  • Specific error messages

  • Configuration issues found

  • Recommended fixes

  • Preventive measures


Things to Know

  • Debug data is available for recent jobs

  • Sensitive information may be redacted

  • Technical details require familiarity with the system

  • Some issues may require escalation to support

Last updated