CodeRush.run home

Chat answers · paste, do not clean up

Paste the answer,
not just the code.

A model hands you prose, a fenced HTML block, a CSS block and a script. Paste all of it into CodeRush and it extracts the fences, combines them into one page, and runs it. No trimming, no stitching, no new file.

  • Several fences combined
  • Keep the whole paste with one click
  • Works with any assistant

What happens when you paste a reply

Extraction, then the normal run.

A paste is a run command here. CodeRush looks at what arrived: if it is a chat answer wrapped around fenced code, the fences are what you meant, so those run and a toast says whether it ran one block or combined several. If the answer was split into markup, styles and a script, they are composed into a single page in that order.

If you wanted the prose instead, the same toast offers Keep the whole paste, which puts the entire reply back in the editor and runs it as Markdown. Pasting over unsaved work is undoable too.

The block a countdown answer usually containsHTML and script
<!doctype html>
<h2>Countdown</h2>
<output id="t">10</output>
<script>
  let left = 10;
  const tick = setInterval(() => {
    t.textContent = --left;
    if (!left) clearInterval(tick);
  }, 1000);
</script>

Run the extracted page

The part chat answers usually get wrong

Missing libraries and missing caution.

Models write new THREE.Scene() and forget the script tag. If a bare script reaches for a library CodeRush knows, it is loaded from cdnjs at a pinned version and the pad tells you which one, so the snippet runs instead of throwing ReferenceError at line one.

Generated code is also code you have not read. The preview is a sandboxed frame with no access to the pad, outside hosts pause for consent before their first request, and Review code before it runs — one command from the palette — holds any page with scripts or connections until you say go. Run offline shows the markup and inline CSS with nothing executing at all.

Then iterate, with or without the model

Your key, your provider, your choice.

The AI panel can take it from there: your prompt and the current source go directly to the provider you chose, with your own key, only when you press Send. CodeRush is not a proxy in that request, and the draft it streams back stays scriptless until it passes the same consent boundary as any other source.

When it works, send it as a Rush link instead of re-pasting the answer into a chat. The recipient gets a page that runs, not a transcript to reassemble.

Questions

Do I have to strip the prose out of the answer first?
No. Paste the whole reply; the fenced code is extracted and run, and a toast offers to keep the whole paste instead.
What if the answer has three separate code blocks?
They are combined into one page — markup, styles and script — and the toast tells you how many blocks it combined.
Does this work with answers from other assistants?
Yes. The extraction looks at fenced code blocks, not at which assistant produced them.
Is it safe to run code a model wrote?
It runs in a sandboxed frame with no access to the pad, and outside connections pause for consent. Turn on Review code before it runs to approve scripts before the first render.

Made out of curiosity.

Paste the next answer and see if it works.