From @peschwa

02​:03 < lucs> m​: say so("a" ~~ / (.) <?{ $0 ~~ /'a'/ && say("\$0 is now ", $0.WHAT) }> /)
02​:03 <+camelia> rakudo-moar 091ee7​: OUTPUT«$0 is now Nil␤True␤»
[...]
02​:05 < lucs> TimToady​: So would that ^^ be a bug or a don't-do-that?
[...]
02​:06 < psch> m​: say so("a" ~~ / (.) <?{ $0 ~~ /'a'/ && say("\$/ is now ", $0.WHAT) }> /)
[...]
02​:06 <+camelia> rakudo-moar 091ee7​: OUTPUT«$/ is now Nil␤True␤»
02​:06 < psch> huh
02​:06 < psch> that seems wrong to me, at least
02​:06 < psch> well, no
02​:07 < psch> accessing $/ there is a "works sometimes" workaround for accessing the CURSOR special name
[...]
02​:07 < psch> m​: say so("a" ~~ / (.) <?{ $0 ~~ /'a'/ &amp;&amp; say("\$¢ is now ", $¢.WHAT) }> /)
02​:07 <+camelia> rakudo-moar 091ee7​: OUTPUT«$¢ is now Nil␤True␤»
[...]
02​:08 < psch> m​: say so("a" ~~ / (.) <?{ $0 ~~ /'a'/ &amp;&amp; say("\$/ is now ", $/.WHAT) }> /)
02​:08 <+camelia> rakudo-moar 091ee7​: OUTPUT«$/ is now (Match)␤True␤»
[...]
02​:08 < psch> m​: say so("a" ~~ / (.) <?{ $0 ~~ /'a'/ &amp;&amp; say("\$/[0] is now ", $/[0].WHAT) }> /)
02​:08 <+camelia> rakudo-moar 091ee7​: OUTPUT«$/[0] is now Nil␤True␤»
02​:08 < psch> lucs​: the hunch is that the switch from Perl6​::Grammar to NQP​::P6Regex messes something up with accessing the different
  interpretations of $/
02​:09 < psch> as in, the code block inside the regex doesn't build the same lookup for $0 as it would outside of a regex
[...]
02​:09 < psch> that might easily be wrong though, but the fact that $/ is Match seems to hint at it...
[...]
02​:11 < psch> 'cause even if $/ isn't supposed to work there (which ISTR it isn't), at least $¢ should

In case of bad editing refer to http://irclog.perlgeek.de/perl6/2015-12-20#i_11746338 and onward.

The underlying issue seems to be that the special name $/ retains some kind of Match object when referenced inside any kind of code assertion inside of a regex, but accessing positional captures fails. I seem to recall mention of $/ not necessarily being valid in a regex, but trying to access $¢ instead is similarly fruitless.