Played with the new Loom SDK streaming helpers to build a log explainer

Paste a build log, get a three-line summary and the likely cause. The streaming helpers made the UI trivial — tokens render as they arrive and the final structured result replaces the stream.

const result = streamObject({
  model,
  schema: z.object({ summary: z.string(), cause: z.string(), fix: z.string() }),
  prompt: `Explain this build log:\n${log}`,
});

Feedback welcome, especially on prompts for very long logs.

For long logs, trim to the last 400 lines plus any line containing error — the model does much better than with the whole thing.