1. Find the request
Open Traces and filter by Agent, Channel, time, or outcome. The CLI can also list recent Traces:
2. Read from top to bottom
Start with Channel input or direct Agent input. Then inspect the model, Gateway, tool or MCP, approval, and output steps.
3. Check the capability
Confirm the capability name and input. If it did not run, check whether the reason was Block, Review rejection, or governance unavailability.
4. Check delivery separately
An Agent may complete successfully while Twilio, Mailgun, or another channel provider fails to deliver the final response. Check the Channel output step separately from Agent execution.
5. Use the CLI for detail
Screenshot placeholder
Add /images/guides/request-tracing.png showing one request with input, model, Gateway, tool, and Channel output steps.
See Traces and Troubleshooting.Last modified on July 27, 2026