Ruby -p, 165 164 159 bytes

It ended up being very similar to the sed answer, but it abuses Ruby's string interpolation to duplicate the hex group matching ([\da-f]{1,2}) within the third regex three times instead of needing to repeat the whole thing again.

• -1 byte from @randomdude999.
• -5 bytes from leveraging @Xcali's Perl solution

gsub /\bPPCG\b/,"CGCC" gsub /\bProgramming Puzzles( \S+ )(Code Golf)\b/,'\2\1Coding Challenges' [1,2].map{|i|gsub /(^|\s)#\K#{'([\da-f]{%d})'%i*3}\b/,'\2\4\3'} 

Try it online!

C++ (gcc), 270 285 283 bytes

Thanks to Neil for pointing out a bug.

-2 bytes thanks to ceilingcat.

#import<regex> #import<string> auto f=[](auto s){typeof(s)R[][2]{"bPPCG","CGCC","bProgramming Puzzles( \\S+ )(Code Golf)","$2$1Coding Challenges","B#(?=([\\da-f]{3}){1,2}\\b)(.+?)(..??)(..??)","#$2$4$3"};for(auto r:R)s=std::regex_replace(s,std::regex('\\'+*r+"\\b"),r[1]);return s;}; 

Try it online!

Retina 0.8.2, 153 130 bytes

\bPPCG\b CGCC \bProgramming Puzzles( \S+ )(Code Golf)\b $2$1Coding Challenges \B(#(?=([\da-f]{3}){1,2}\b).+?)(..??)(..??)\b $1$4$3 

Try it online! Link includes test cases. All substitutions are case sensitive. Assumes normal regex word characters are acceptable so that \B# matches only #s that don't follow a word character. Edit: Saved 22 bytes thanks to @tsh.

GNU sed -E, 198 chars

s/\bPPCG\b/CGCC/g s/\bProgramming Puzzles( \S* Cod)e Golf\b/Code Golf\1ing Challenges/g s/((^|\W)#[0-9a-f])([0-9a-f])([0-9a-f])\b/\1\4\3/g s/((^|\W)#[0-9a-f]{2})([0-9a-f]{2})([0-9a-f]{2})\b/\1\4\3/g 

Runnable using e.g. sed -E 'the above'; newlines can either be included literally, or replaced with ; if you so wish. Both work.

@HyperNeutrino come on that word boundary rule is stupid. Look what I had to do in the # case.

Yes, I didn't even try. :P

+9 by @Value Ink

05AB1E, 123 109 105 110 114 bytes

žKISå_Å¡JεÐć'#QsžhA6£«sSåP*i3äćsRJ«ë"PPCG"Qi"CGCC"]J”–±ÇÀ”DU¡ćsε”ƒËŠˆ”©¡DVćDÁ2ôεðå}ćs_P*YyÊP*i”Âïªï”«s®ý«®ìëyXì]J« 

+5 bytes fixing test cases like Programming Puzzles and no substitution Code Golf and Programming Puzzles and Programming Puzzles and Code Golf.
+4 bytes fixing test cases like color-#00f (colors with something besides spaces/newlines surrounding it). Thanks to @Grimy for bringing this to my attention.

Case-sensitive. Hexadecimal values are with lowercase abcdef; Programming Puzzles ... Code Golf is in titlecase; PPCG is in full uppercase.

Try it online.

Definitely not the right language for the job.. Mimicking word-boundaries and replacing Programming Puzzles \S+ Code Golf, but not Code Golf \S+ Programming Puzzles or Programming Puzzles \S+ \S+ Code Golf without any regexes is pretty hard (to do short).. >.>

Explanation:

žK # Push "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789" # (NOTE: if '_' also count as word boundary, use `žj` instead) IS # Push the input as a list of characters å_ # Check for each character if they are NOT in the string Å¡ # Split the (implicit) input on truthy values J # Join each inner character list to a string again ε # Map each word to: Ð # Triplicate the word ć # Extract head; push remainder and head separately to the stack '#Q '# Check if the head equals "#" žh # Push string "0123456789" A6£« # Append the first 6 letters of the alphabet: "0123456789abcdef" s sSåP # Check if the characters of the remainder are all in this string *i # If both are truthy: 3ä # Split the 'word' into three parts # i.e. "#ab12cd" → ["#ab","12","cd"] ćs # Extract head, and swap R # Reverse the remainder list J« # Join them together, and merge them to the head again ë"PPCG"Qi # Else-if the word is "PPCG": "CGCC" # Push "CGCC" # (Implicit else: # Use the word that's still there from the initial triplicate) ] # Close all if statements and the nested map J # Join the mapped words together again ”–±ÇÀ” # Push dictionary string "Programming Puzzles" DU # Save a copy in variable `X` ¡ # Split the string by this ćs # Extract head & swap; pushing head and remainder to the stack ε # Map each substring `y` in the remainder to: ”ƒËŠˆ” # Push dictionary string "Code Golf" © # Save it in variable `®` (without popping) ¡ # Split the current substring we're mapping by it DV # Save a copy of this list in variable `Y` ćD # Extract the head and duplicate # (so the stack is: remainder, head, head) Á # Rotate the characters in the head-string once towards the right 2ô # Split it into parts of size 2 εðå} # Check in each part if it contains a space ćs # Extract head and swap again _ # Check if all values are 0 P # And check if this is truthy for all * # And check if this is truthy, as well as the head # (this means the current string has a leading and trailing space, # and no other spaces) YyÊP # Check that none of the susbtrings in variable `Y` # are equal to the current substring `y` *i # If both checks above are truthy: ”Âïªï”« # Prepend "Coding Challenges" to the duplicated head s®ý # Join the remainder by variable `®` ("Code Golf") « # Append it ®ì # And prepend an additional variable `®` ("Code Golf") ë # Else: y # Simply keep the substring `y` as is Xì # And prepend variable `X` ("Programming Puzzles") # since we've split by it ] # Close all if-else statements and the map J # Join the mapped substrings together to a single string « # And append it to the initially extracted head # (then output the resulting string implicitly as result) 

See this 05AB1E tip of mine (section How to use the dictionary?) to understand why ”–±ÇÀ” is "Programming Puzzles"; ”ƒËŠˆ” is "Code Golf"; and ”Âïªï” is "Coding Challenges".

Python 2, 240 bytes

import re lambda x,s=re.sub,b='(?<!\w)',e='(?!\w)',h='([\da-f]',t=r'#\1\3\2':s(b+'#%s{2})'%h+h+'{2})%s{2})'%h+e,t,s(b+'#%s)'%h+h+')%s)'%h+e,t,s(b+'Programming Puzzles( \S+ Cod)e Golf'+e,r'Code Golf\1ing Challenges',s(b+'PPCG'+e,'CGCC',x)))) 

Try it online!

JavaScript (Node.js), 174 bytes

s=>s[R='replace'](/\bPPCG\b/g,'CGCC')[R](/\bProgramming Puzzles( \S+ )(Code Golf)\b/g,'$2$1Coding Challenges')[R](/\B#(?=([\da-f]{3}){1,2}\b)(.+?)(..??)(..??)\b/ig,'#$2$4$3') 

Try it online!

Pyth, 177 173 162 142 bytes

J::jb.z"\\bPPCG\\b""CGCC"." z¶NZI°Pÿúd(MÜ_BöIkxnqä'u)"." s6#~ÍN³=<nñu/GÎg"VS2~:J%"(^|\W)#%s\\b"*3%"([\da-f]{%d})"N$r"\1#\2\4\3"$)J 

Here's a version without Pyth's string compression mechanisms (aka it's safe to copy-paste):

J::jb.z"\\bPPCG\\b""CGCC""\\bProgramming Puzzles( \S+ )(Code Golf)\\b""\\2\\1Coding Challenges"VS2~:J%"(^|\W)#%s\\b"*3%"([\da-f]{%d})"N$r"\1#\2\4\3"$)J 

Try it online!

This ended up being really long because I tried to be as pedantic as possible with the regexes. I tried to compress every string possible, but most of them either didn't get smaller or couldn't be pasted into TIO properly.

Explanation:

J:: # definition of J to the following 2 regex replacements jb.z # input to first regex replacement: all input lines joined together "\\bPPCG\\b" # first regex "CGCC" # first replacement ."<compressed>" # second regex: "\\bProgramming Puzzles( \S+ )(Code Golf)\\b" ."<compressed>" # second replacement: "\\2\\1Coding Challenges" VS2 # loop twice, N = 1 or 2 ~:J # some weird assignment: J = regex replace in J (would be J := (regex, replacement) if : was python's regex replace operator) % # third regex: string format "(^|\W)#%s\\b" # format string *3 # repeat 3 times: %"([\da-f]{%d})"N # string format, replace %d with N (the loop counter) $r"\1#\2\4\3"$ # third replacement: uses python raw literals because it's shorter than escaping the backslashes ) # end for loop J # print J 

• -11 thanks to a better regex from Value Ink's Ruby answer
• -20 thanks to using a loop for both of the hex replacements, inspired by the Ruby and Perl answers

Perl 5 -p, 152 145 bytes

@ValueInk saves 7 bytes

s/\bPPCG\b/CGCC/g;s/\bProgramming Puzzles( \S+ )(Code Golf)\b/$2$1Coding Challenges/g;for$a(1,2){$p="([a-f0-9]{$a})"x3;s/(^|\s)#\K$p\b/$2$4$3/gi} 

Try it online!

Java 8, 192 bytes

s->s.replaceAll("\\bPPCG\\b","CGCC").replaceAll("\\bProgramming Puzzles( \\S+ )(Code Golf)\\b","$2$1Coding Challenges").replaceAll("\\B(#(?=([0-9a-f]{3}){1,2}\\b).+?)(..??)(..??)\\b","$1$4$3") 

Port of @Neil's Retina answer, so make sure to upvote him!

Try it online.