Scenario is to create tgz archives on a remote server (tar -czf --to-stdout directory).
Currently, captureing the output will be done in memory which is not feasible for large output. Implementation could be specific for files (like capture %Q{tar czf -O}, into_file: 'bigarchive.tgz') or preferrably more general (capture %Q{}, into_io: File.open(...)).
With a bit of headwrapping, this could also be modelled as download_cmd_out tar_cmd, to: 'stdout.file'.
Note that while i refer to tar commands, this might be useful for other commands that produce a lot of data, too.