What this week is
Your agent can build things — now let it show them, and run untrusted code without risking your machine.
This week you add artifact rendering and a live-preview server so a generated page or chart appears in your browser and refreshes as the agent tweaks it; a background-task runner for long jobs; and a real secure sandbox — OS-enforced limits, a scrubbed environment, a directory jail.
The sandbox is the week's real lesson: model-written code is untrusted code, and the only durable answer is enforcement by the operating system, not by a prompt.
The walkthrough
Each part adds one capability to the same growing codebase, and each is a runnable checkpoint.
- 01ArtifactsRender what the agent produces instead of printing it.
- 02The preview serverLive reload so you watch the agent iterate in real time.
- 03The sandboxOS-enforced limits, a scrubbed environment, and a directory jail.
- 04Background tasksLong jobs that run without blocking the conversation.
After this week you can
- Render agent output as live, previewable artifacts
- Sandbox untrusted code with real OS enforcement
- Run long tasks without freezing the interface