A middle ground between the full JSON output and the human-focused plain output. My idea here is integrating reuse lint with tools like reviewdog where it excels with output along the lines of:

scripts/foo1.sh: missing license 'MPL-2.0'
scripts/foo2.sh: no licensing information
scripts/foo3.sh: no copyright information

One can generate such an output using the JSON version and processing it, but I think it would be beneficial to include it upstream.