(It affected commands on separate lines too I think - I just used semicolons to save space in describing the problem)

false | true (false pipe to true) should cause failure - due to set -o pipefail.

false || true (false or true) should continue to work, however.

For the record, on master this does not get trapped either (i.e. the single-line semi-colon construct isn't relevant - so I'll change the issue title slightly):

[runtime]
   [[foo]]
       script = """
true
false
true"""

Actually, on master even an error in an environment var definition will not be trapped.

A bit surprised by that. How did scripts that were expected to fail worked in the test battery? Do we by chance only have command failure at the end of functions?

I guess so! I've just made a new test that confirms this (passes here, fails on master). I'll put up a PR.