Don't know what to tell you other than try providing a SSCCE that can actually be run. This below works as I expect it so you're doing something strange or (entirely possible) your problem statement's being misread.

(Also the <code> formatting is doing weird things but I actually have literal ✶ in my source and the output where it's being replaced with the entity below everywhere save the initialization of $wonky_char. Not sure what's the right way to get literal UTF8 chars in sample code using utf8.)

#!/usr/bin/env perl
use 5.034;

use HTML::Entities qw( decode_entities );
use utf8;

my $input = qq{{✶Adjektive (Nominalflexion)~87✶}};
my $wonky_char = decode_entities( q{✶} );

binmode( STDOUT, q{:utf8} );

say qq{\$input: $input};
say qq{\$wonky_char: $wonky_char};

my $to_match = "Adjektive (Nominalflexion)~87";
my $new_string =
  $input =~ s{\{$wonky_char(\Q$to_match\E)$wonky_char\}}{<div>I found 
+'$1'</div>}r;

say qq{\$new_string: $new_string};

my $cleaner_regex_sample =
  $input =~ s{ \{ $wonky_char (\Q$to_match\E) $wonky_char \} }{<div>Al
+so found '$1'</div>}rx;

say qq{cleaner: $cleaner_regex_sample};

exit 0;

__END__
$input: {&#10038;Adjektive (Nominalflexion)~87&#10038;}
$wonky_char: &#10038;
$new_string: <div>I found 'Adjektive (Nominalflexion)~87'</div>
cleaner: <div>Also found 'Adjektive (Nominalflexion)~87'</div>
[download]