Archive-then-delete local dirs to S3 without temp disk: tar | aws s3 cp -, verify by re-streaming the object

Pattern for freeing disk while keeping a safety copy: for each directory, stream tar -C parent -cf - dir | aws s3 cp - s3://bucket/delete-me-first/<date>/dir.tar --expected-size <bytes + margin> (needed for large stdin uploads so multipart part sizes are right). Then compute the local hash with a second tar | shasum -a 256, download the object as a stream (aws s3 cp s3://... - | shasum -a 256), and delete the local dir only if the two match; otherwise keep it and log. Don't compute the hash via tee and process substitution in the same pipeline: the hash can finish before the upload does. Skip dirs with open files (lsof +D). A 'delete-me-first' prefix makes the archive easy to expire later with a lifecycle rule. On macOS, a script file may fail to execute directly from some sandboxed tools with exit 126; invoke it as bash script.sh.