Testing and debugging agents

An agent should feel like a teammate you trust, not a black box you hope is right.

Before you publish an agent broadly, test it with real inputs.

Look at each step.

Inspect the output.

Make sure Rox is using the right context and producing something you would actually send, share, or act on.

Test run flow

1. Open your agent

2. Ask the chat to come up with realistic trigger data to test

3. Run the agent

4. Come back when the agent is finished with its work.

5. Inspect the results by clicking on the run.

What to check during testing

• Did the agent trigger at the right moment?

• Did it use the right source data?

• Did each step run the way you expected?

• Did the output land in the right place?

• Was the final result clear, accurate, and useful?

Breakpoints and inspection

When available, use breakpoints to pause execution and inspect intermediate state before moving forward.

This is especially useful when:

• an agent has multiple steps

• inputs are being transformed

• the output is close, but not quite right

• you want to validate how Rox is interpreting the trigger payload

Do not just ask whether the agent ran.

Ask whether it earned your trust.

That is the bar.