# Replicate: Smoke test, built by GPT-6 Luna

From **Prism** (https://prism.berrocal.xyz/i/zz-smoke/gpt-6-luna/). Six models got the same brief; this is what **GPT-6 Luna** (`gpt-6-luna` in Codex CLI 0.158.0, effort `xhigh`) built autonomously in 1m 34s. No human edited it.

License: **CC0-1.0**, a public-domain dedication. Use it for anything.

## Get the exact files (use curl, not a page fetcher: fetchers summarize and truncate)

```bash
mkdir -p zz-smoke-gpt-6-luna && cd zz-smoke-gpt-6-luna
curl -fsSLO https://prism-artifacts.berrocal.xyz/zz-smoke/gpt-6-luna/pack.zip
unzip -q pack.zip && shasum -a 256 -c SHA256SUMS
```

`SHA256SUMS` paths are relative to the folder you unzipped into, so run the check from inside it. The website is `site/` with entry `site/index.html`. Preview it with any static server, e.g. `npx serve site`. It is plain static files, so it works under any sub-path.

## Three ways to use it

1. **Clone:** copy `site/` as-is. It is byte-exact; `SHA256SUMS` proves it.
2. **Adapt:** rebuild it inside this project. Read `site/index.html` (and any other files in `site/`) plus `sources/NOTES.md`, keep the art direction, interactions and generated resources, and re-implement them using this project's framework and conventions.
3. **Re-roll:** give your own model the exact brief in `sources/PROMPT.md` and compare your result with the six on Prism.

## What's in the pack

- `site/`: the artifact, 2 files, 93 KB
  - `site/beep.wav` (88 KB)
  - `site/index.html` (4 KB)
- `sources/PROMPT.md`: the exact prompt every model received (brief plus house rules)
- `sources/NOTES.md`: the model's own notes on what it built
- `sources/workspace/`: anything else the model wrote outside `site/`, such as scripts that generated the resources
- `sources/transcript.json`: the full agent session, normalized; `transcript.jsonl` is the raw event stream
- `input/`: the input resources that came with the brief, if any
- `pack.json`: the manifest, with a sha256 and MIME type for every file

## Run stats

| | |
|---|---|
| Model | GPT-6 Luna (`gpt-6-luna`) |
| Harness | Codex CLI 0.158.0, effort `xhigh` |
| Wall time | 1m 34s (cap 8 min) |
| Tool calls | 5 |
| Output tokens | 3,929 |
| Self-checks run | 1 |

Browse every model's version of this brief at https://prism.berrocal.xyz/i/zz-smoke/.