When use ETL Adapter CSV to write flow with the AwsS3Stream
the csv column headers is repeated every batch size

I think it's this line :
https://github.com/flow-php/etl-adapter-csv/blob/230d8e40de9e26b7e7726bc6ee2fe395104c4255/src/Flow/ETL/Adapter/CSV/CSVLoader.php#L97

in case of aws s3 the file not exist during batch process (real write only at end)

Perhaps use
$context->streams()->isOpen
instead of $context->streams()->exists