Re. failures with tap 14, see https://github.com/richardlau/node-inspect/runs/401645634 for example failures. Most of them were timeouts. I haven't dug into them here (due to my unfamiliarity with node-inspect) but in node-report I did find that tap 14 interfered with its tests (nodejs/node-report#135) by installing signal handlers which may be the issue here if signals are being used.

The failures with tap 12 and Node.js 10 seem to be off-by-one line mismatches?
e.g. https://github.com/nodejs/node-inspect/pull/78/checks?check_run_id=402008893#step:4:47

        not ok 3 - pauses in next line of the script
          ---
          found: |-
            break in examples/break.js:1
            > 1 const x = 10;
              2 let name = 'World';
              3 name = 'Robin';
            debug> 
          pattern: 'break in examples/break.js:2'
          at:
            line: 31
            column: 9
            file: test/cli/break.test.js
            function: cli.waitForInitialBreak.then.then.then.then
          stack: |
            cli.waitForInitialBreak.then.then.then.then (test/cli/break.test.js:31:9)
          source: |
            t.match(
          ...

From #75 (comment)

It may be worth it to bump the major version and set the minimum version to 12. I don't think we'd back-port node inspect changes to node 10 given how infrequent these were in general.

@jkrems I could always leave 10.x out of this PR, which would allow a passing CI on 12 and 13 (which would be a better position than no CI at all). We could then add 10.x in a later PR if deemed necessary.

Rebased on master and commented out 10.x. Thanks for setting this up, this is great!