Error handling

[DV-Jonas]

@yusuf-eren How would you recommend best handling errors? I have a agent with a structured output that keeps getting a maxTurns error. It looks to be the agent validator that keeps invalidating the response (invalid json likely). ModelBehaviorError is never thrown and maxTurns Error doesn't have any usable information. I've enabled debug logging but also not getting anything useful from that.

[DV-Jonas]

I realize I should be using the tracing functionality. Would it be possible for you to add some examples on how to use the different tracing functionality?

[yusuf-eren]

Hi @DV-Jonas , Happy to see you using library! And also thanks for creating another issue for the library. It keeps me motivated to continue to this journey!

For tracing, there is nothing to set actually. I added tracing support to the library at v.0.1.8 so make sure to upgrade the package(I recommend you to upgrade to 0.1.9 by the way).
If it's already upgraded;

Go to https://platform.openai.com/traces and check for your agent's logs. It sets the workflow name to Agent workflow by default. But if you want to set a custom workflow name you can pass it under the runConfig

Here is an example;

import { Agent, RunConfig, Runner } from 'openai-agents-js';

async function main() {
    const agent = new Agent({
        name: 'My Agent',
        instructions:
            'You are a helpful assistant that can answer questions and help with tasks.'
    });

    const result = await Runner.run(agent, 'How many days in a year?', {
        runConfig: new RunConfig({
            workflowName: '[openai-agents-js]: Hello World!'
        })
    });

    console.log(result.finalOutput);
}

main();

Tracing Output

"How to debug the issue?"

You need to click to the trace you just run and see the output of each call. There may be a tool call gives the same response again and again so it reaches to 10 maximum steps and stucks. 10 is the default value for maxTurns. If you have a long running agent feel free to increase that value in the Runner parameters;

 const result = await Runner.run(agent, 'How many days in a year?', {
      maxTurns: 50,
      runConfig: new RunConfig({
          workflowName: '[openai-agents-js]: Hello World!'
      })
  });

If you have any other questions or are stuck with it please let me know so we can dive into it.