1. Click into a run from the run log
Activity feed
2. The left hand pane displays an activity feed of all agent actions in chronological order
3. Click on any action to see the corresponding screenshot in the preview pane
Preview
4. You can also navigate via the scrubber and arrows in the preview pane, which displays screenshots of the agent's view
Agent navigator
5. Click into the filter icon in the activity feed to see additional information and controls
6. The [[Agent navigator]] is a hierarchical representation of the run - see how and when each agent was called as well as a status for each agent (e.g. complete, incomplete, running, etc.)
7. Click on the arrow icon to jump straight to an agent's first action
8. Click on one or more rows then click "Apply" to filter the activity feed by the selected agents only
9. Click the "Show values" toggle to also view the inputs and outputs for each agent
10. To remove all filtering, click "Clear"
11. Then click "Apply"
Agent details
12. The third pane located at the bottom of the page displays the agent's configuration (e.g. model, tools, inputs/outputs, etc.)
13. Scroll further down this section to view the agent's prompt
Dev mode for advanced users
14. Advanced users may opt into Dev mode by clicking on their profile, then clicking on the "Dev mode" toggle
15. While Dev mode is on, inputs and outputs will be displayed in JSON format
16. Clicking on an action will display additional context such as the exact tool the agent used and relevant metadata
17. Click the "View rendered prompt" toggle to see the prompt exactly as it was shown to the agent
