Return to Answer

PowerShell, 108 100 95 Bytes

$i,$z=$args;$b,$s=[char[]]$z;($b+(($i-split'[^\d.-]+'-ne'')-join$s)+' }) >]'[($b-32)%6]).Trim()

A slight variation on the below, thanks to TessellatingHeckler. This takes the two arguments as input to $i and $z, then casts $z as a char-array and pulls the two arguments out into $b and $s. We then concatenate $b with $i having been split on digits and negatives that exist (i.e., -ne''). That gets -joined with the same indexing, and finally .Trim() to remove leading/trailing spaces.

###Old:

param($a,$b)($b[0]+(-split($a-replace"[\[\]{}()<>;,]",' ')-join$b[1])+' }) >]'[($b[0]-32)%6]).Trim()

Pretty similar to the other answers, just with PowerShell's verbose unique syntax.

###Explained:

• Takes input as two strings via param($a,$b), then builds the output string via everything inside the parens ().
• We first use the first character of $b with $b[0], add that to another parens-separated dynamically built string, and then add that to the closing character.
• The closing character is selected by indexing into a string based on a formula of ($b[0]-32)%6 - thanks to Danko Durbić for golfing 8 bytes here).
• The inner-parens string is constructed by simply getting rid of any input junk with a regex and -replaceing it with a space, -splitting that on spaces to create an array, and -joining that array together with the second character of $b.
• Finally, the trailing .Trim() removes any remaining leading or trailing spaces.

PowerShell, 108 100 95 85 Bytes

$i,$z=$args;($z[0]+($i-split'[^\d.-]+'-ne''-join$z[1])+' }) >]'[($z[0]-32)%6]).Trim()

_{(see revision history for previous versions)}

Golfed another 15 bytes by removing $b and $s variables and changing parens on the inner string. 

This takes input as two strings and stores them into $i and $z, then we construct a new output string. The inner parens -splits $i with a regex to select only numeric digits, then -joins back together with the requested delimiter. We concatenate that with the first character of the delimiter input (e.g., [) and close it with indexing into a string based on the ASCII value of the first character and some formulation trickery. The outer .Trim() removes any leading or trailing spaces.

PowerShell, 108 100 Bytes

param($a,$b)($b[0]+(-split($a-replace"[\[\]{}()<>;,]",' ')-join$b[1])+' }) >]'[($b[0]-32)%6]).Trim()

Pretty similar to the other answers, just with PowerShell's verbose unique syntax.

###Explained:

• Takes input as two strings via param($a,$b), then builds the output string via everything inside the parens ().
• We first use the first character of $b with $b[0], add that to another parens-separated dynamically built string, and then add that to the closing character.
• The closing character is selected by indexing into a string based on a formula of ($b[0]-32)%6 - thanks to Danko Durbić for golfing 8 bytes here).
• The inner-parens string is constructed by simply getting rid of any input junk with a regex and -replaceing it with a space, -splitting that on spaces to create an array, and -joining that array together with the second character of $b.
• Finally, the trailing .Trim() removes any remaining leading or trailing spaces.

PowerShell, 108 100 95 Bytes

$i,$z=$args;$b,$s=[char[]]$z;($b+(($i-split'[^\d.-]+'-ne'')-join$s)+' }) >]'[($b-32)%6]).Trim()

A slight variation on the below, thanks to TessellatingHeckler. This takes the two arguments as input to $i and $z, then casts $z as a char-array and pulls the two arguments out into $b and $s. We then concatenate $b with $i having been split on digits and negatives that exist (i.e., -ne''). That gets -joined with the same indexing, and finally .Trim() to remove leading/trailing spaces.

###Old:

param($a,$b)($b[0]+(-split($a-replace"[\[\]{}()<>;,]",' ')-join$b[1])+' }) >]'[($b[0]-32)%6]).Trim()

Pretty similar to the other answers, just with PowerShell's verbose unique syntax.

###Explained:

• Takes input as two strings via param($a,$b), then builds the output string via everything inside the parens ().
• We first use the first character of $b with $b[0], add that to another parens-separated dynamically built string, and then add that to the closing character.
• The closing character is selected by indexing into a string based on a formula of ($b[0]-32)%6 - thanks to Danko Durbić for golfing 8 bytes here).
• The inner-parens string is constructed by simply getting rid of any input junk with a regex and -replaceing it with a space, -splitting that on spaces to create an array, and -joining that array together with the second character of $b.
• Finally, the trailing .Trim() removes any remaining leading or trailing spaces.

PowerShell, 108 Bytes

param($a,$b)($b[0]+(-split($a-replace"[\[\]{}()<>;,]",' ')-join$b[1])+"]})>"["[{(< ".IndexOf($b[0])]).Trim()

Pretty similar to the other answers, just with PowerShell's verbose unique syntax.

###Explained:

• Takes input as two strings via param($a,$b), then builds the output string via everything inside the parens ().
• We first use the first character of $b with $b[0], add that to another parens-separated dynamically built string, and then add that to the closing character (which is selected by indexing into a string based on the index of $b[0] into another string).
• We exploit the fact that strings can be indexed like an array, but without triggering "Array index out of bounds" exceptions, so that the .IndexOf() doesn't return -1 if the "bracket" character is a space. That's why the "closing brackets" string is length-4, while the "indexing" string is length-5.
• The inner-parens string is constructed by simply getting rid of any input junk with a regex and -replaceing it with a space, -splitting that on spaces to create an array, and -joining that array together with the second character of $b.
• Finally, the trailing .Trim() removes any remaining leading or trailing spaces.

PowerShell, 108 100 Bytes

param($a,$b)($b[0]+(-split($a-replace"[\[\]{}()<>;,]",' ')-join$b[1])+' }) >]'[($b[0]-32)%6]).Trim()

Pretty similar to the other answers, just with PowerShell's verbose unique syntax.

###Explained:

• Takes input as two strings via param($a,$b), then builds the output string via everything inside the parens ().
• We first use the first character of $b with $b[0], add that to another parens-separated dynamically built string, and then add that to the closing character.
• The closing character is selected by indexing into a string based on a formula of ($b[0]-32)%6 - thanks to Danko Durbić for golfing 8 bytes here).
• The inner-parens string is constructed by simply getting rid of any input junk with a regex and -replaceing it with a space, -splitting that on spaces to create an array, and -joining that array together with the second character of $b.
• Finally, the trailing .Trim() removes any remaining leading or trailing spaces.