The results are already somewhere

Every pipeline on an ephemeral sandbox ends the same way: in the final seconds before teardown, push the outputs to a bucket, because the filesystem is about to stop existing.

On Hoody, the filesystem doesn't stop existing. The reports your agent generated are sitting in /output in a container that's still there. hoody-files serves that directory over HTTP: each file is a URL, and ?zip on the directory turns it into an archive. Your product hands the user a link into the machine that did the work.

No bucket, no upload step, no lifecycle policy on the bucket you didn't create. The artifact store is the sandbox's own disk, addressable from birth.

The last step of every job was "upload the results." It's gone.