Agent SDK launch
Ground “agents as code” in configuration, execution, and operations. Code establishes control; the run dashboard demonstrates ongoing visibility of results.
View HTML source
<!doctype html><html lang="zh-CN"><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>Agent SDK launch · sequence</title><style>html,body{margin:0;width:100%;height:100%;overflow:hidden;background:#f5f5f1}canvas{display:block;width:100%;height:100%;object-fit:contain}</style><body data-mode="sequence"><canvas width="1000" height="600"></canvas><script src="data.js"></script><script src="common.js"></script><script src="scene.js"></script><script src="bridge.js"></script></body></html>Motion and composition
Local motion · Composition rules
Show minimal configuration before run status
After the opening heading, show agent files or a task directory, change build status to deployed, then enter the run dashboard. Retain the task name at each transition to connect code and results.
Support technical information with a calm layout
Use simple sans serif on white explanatory pages, covering one capability per sentence. Use monospace for code or filenames and aligned dashboard numerals; confine warm backgrounds outside product windows, not beneath text.
Reserve accents for selected tasks, current status, or key counts; label success and running states explicitly rather than relying on color.
Leave room for an improvement action after running
View HTML source
<!doctype html><html lang="zh-CN"><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>Agent SDK launch · sequence</title><style>html,body{margin:0;width:100%;height:100%;overflow:hidden;background:#f5f5f1}canvas{display:block;width:100%;height:100%;object-fit:contain}</style><body data-mode="clip" data-start="15" data-end="27"><canvas width="1000" height="600"></canvas><script src="data.js"></script><script src="common.js"></script><script src="scene.js"></script><script src="bridge.js"></script></body></html>Keep task names, times, and states visible on the dashboard, focusing first on statistics and then one task. Enter a new scanning or improvement request to close the configuration→run→adjustment loop.
Change tasks, preserve the control relationship
Adapt to design checks, research organization, or site inspections. Update code configuration, task list, evaluation items, and the next request together; each headline claim must map to a later interface action.
Sections and rhythm
Click a section to seek to it in the choreography example above.






Visual references · Discrete still frames
Creating other themes
- Suitable for developer platforms; retain configuration, execution, status, and improvement as four object types
- Replace code, metrics, and task purpose for the target product
Production constraints
Code directory / status bar / deployment feedback / run count / task list / evaluation input
- Follow abstract messages about code configuration with a running dashboard to demonstrate process observability
- transition from building to deployment complete before showing multitask status counts
- return from the dashboard to a new request at the end, completing a run → evaluate → iterate loop