I guess what you mean is this: When using hmmsearch for the Pkinase domain from PFAM against SRC, the output is:

>> SRC
   #    score  bias  c-Evalue  i-Evalue hmmfrom  hmm to    alifrom  ali to    envfrom  env to     acc
 ---   ------ ----- --------- --------- ------- -------    ------- -------    ------- -------    ----
   1 !  169.7   0.0   4.1e-54   4.1e-54       3     254 ..     272     515 ..     270     522 .. 0.88
  Alignments for each domain:  
  == domain 1    score: 169.7 bits;  conditional E-value: 4.1e-54
              EEEEEE----EEEEEEEETTCCEEEEEEEEEGGGHHHHHHHHHHHHHHHHHHHH-TTB--EEEEEEESTEEEEEEE--SC-EHHHHHH..HHCCB-HHHHHH CS
  pkinase   3 lleklGsGsfGkVykakkkktgkkvAvKilkkeeekskkektavrElkilkklsHpnivkllevfeekdelylvleyveggdlfdllk..kegklseeeikk 102
              l+ klG+G fG+V+ +++++t + vA+K+lk  + + ++  +   E++++kkl+H+ +v+l++v+ +++ +y+v+ey+ +g+l d+lk  + + l+  ++ +
      SRC 272 LEVKLGQGCFGEVWMGTWNGTTR-VAIKTLKPGTMSPEAFLQ---EAQVMKKLRHEKLVQLYAVV-SEEPIYIVTEYMSKGSLLDFLKgeTGKYLRLPQLVD 368
              6779*****************96.************999755...*******************9.6889******************9888899******* PP

              HHHHHHHHHHHHHHTCEE-S--SGGGEEEETT--EEE------EEE.CEESS-CSS----GGGS-HHHHHTSHHHHHHHHHHHHHHHHHHHHC.---TTTST CS
  pkinase 103 ialqilegleylHsngiiHrDLKpeNiLldkkgelkiaDFGlakkl.eksseklttlvgtreYmAPEvllkakeytkkvDvWslGvilyellt.gklpfsge 202
              +a qi++g++y+ +++ +HrDL  +NiL+ ++ ++k+aDFGla+ +  ++++  +      ++ APE  l     t k+DvWs+G++l el+t g++p++g
      SRC 369 MAAQIASGMAYVERMNYVHRDLRAANILVGENLVCKVADFGLARLIeDNEYTARQGAKFPIKWTAPEAAL-YGRFTIKSDVWSFGILLTELTTkGRVPYPGM 469
              ********************************************764444445555566679********.999*****************99678999999 PP

              HHHHHHHHHCCCHHTTS--GGGGCSHHHHHHHHHHHHHT-SSGGGS--HHHH CS
  pkinase 203 seedqleliekilkkkleedepkssskseelkdlikkllekdpakRltaeei 254
               +++ l  +e+  +     ++p +++  e+l dl++++++k+p++R+t+e +
      SRC 470 VNREVLDQVERGYRM----PCPPECP--ESLHDLMCQCWRKEPEERPTFEYL 515
              555555555555554....4454544..89******************9865 PP

That is, it starts with residue 3. When you use hmmalign --alllcol, you see that hmmer doesn't want to include the leucine and arginine at the start:

# STOCKHOLM 1.0

SRC          mgsnkskpkdasqrrrslepaenvhgagggafpasqtpskpasadghrgp
#=GR SRC PP  **************************************************
#=GC PP_cons ..................................................
#=GC RF      ..................................................

SRC          saafapaaaepklfggfnssdtvtspqragplaggvttfvalydyesrte
#=GR SRC PP  **************************************************
#=GC PP_cons ..................................................
#=GC RF      ..................................................

SRC          tdlsfkkgerlqivnntegdwwlahslstgqtgyipsnyvapsdsiqaee
#=GR SRC PP  **************************************************
#=GC PP_cons ..................................................
#=GC RF      ..................................................

SRC          wyfgkitrreserlllnaenprgtflvresettkgayclsvsdfdnakgl
#=GR SRC PP  **************************************************
#=GC PP_cons ..................................................
#=GC RF      ..................................................

SRC          nvkhykirkldsggfyitsrtqfnslqqlvayyskhadglchrlttvcpt
#=GR SRC PP  **************************************************
#=GC PP_cons ..................................................
#=GC RF      ..................................................

SRC          skpqtqglakdaweipreslr--LEVKLGQGCFGEVWMGTWNGTTR-VAI
#=GR SRC PP  *******************75..6779*****************96.***
#=GC PP_cons .......................6779*****************96.***
#=GC RF      .....................xxxxxxxxxxxxxxxxxxxxxxxxxxxxx

SRC          KTLKPGTMSPEAFLQ---EAQVMKKLRHEKLVQLYAVV-SEEPIYIVTEY
#=GR SRC PP  *********999755...*******************9.6889*******
#=GC PP_cons *********999755...*******************9.6889*******
#=GC RF      xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

SRC          MSKGSLLDFLKgeTGKYLRLPQLVDMAAQIASGMAYVERMNYVHRDLRAA
#=GR SRC PP  ***********9888899********************************
#=GC PP_cons ***********..88899********************************
#=GC RF      xxxxxxxxxxx..xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

SRC          NILVGENLVCKVADFGLARLIeDNEYTARQGAKFPIKWTAPEAAL-YGRF
#=GR SRC PP  *******************764444445555566679********.999*
#=GC PP_cons *******************76.444445555566679********.999*
#=GC RF      xxxxxxxxxxxxxxxxxxxxx.xxxxxxxxxxxxxxxxxxxxxxxxxxxx

SRC          TIKSDVWSFGILLTELTTkGRVPYPGMVNREVLDQVERGYRM----PCPP
#=GR SRC PP  ****************99678999999555555555555554....4454
#=GC PP_cons ****************99.78999999555555555555554....4454
#=GC RF      xxxxxxxxxxxxxxxxxx.xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

SRC          ECP--ESLHDLMCQCWRKEPEERPTFEYL------qafledyftstepqy
#=GR SRC PP  544..89******************8765......56889999*******
#=GC PP_cons 544..89******************8765.....................
#=GC RF      xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx...............

SRC          qpgenl
#=GR SRC PP  ******
#=GC PP_cons ......
#=GC RF      ......
//

So one thing you could do to force your protein into the domain model is to count the "-" at the beginning and the end and include the correct number of residues.

Problem: extract just the 255-aa protein kinase domains from a set of 70 sequences.

You don't really need to use the split() method; slicing the original sequence should do it once you've calculated the sequence coordinates.

If you're already parsing some of HMMer's plain-text output, you can look at the fields "hmmfrom" and "hmm to" to see how many positions are missing from the start and end of the sequence, then look at the "alifrom" and "ali to" fields to determine the positions of the original sequence that matched the HMM profile. Then, simple arithmetic to get the coordinates in the original sequence:

realstart = alifrom - hmmstart
realend = ali_to + (255 - hmm_to)
realdomain = rec_mine[realstart:realend]

The sequence coordinates of the aligned region also appear in the Stockholm output, in the sequence identifier after the slash character -- so you can also count the number of terminal dashes, minus any leading or trailing insert (lowercase) characters.

Keep in mind that for some purposes, the missing sequence positions might not be meaningful -- e.g. many biologically functional kinases are missing one or more subdomains, so the sequence portions that were left out by the HMM profile could be phylogenetically or structurally unrelated to the typical kinase subdomain. Though if the missing portions are only 3-4 aa I think you're fine, it's just HMMer being stubborn.

I dunno much about HMMER but you could attempt a biopython solution along these lines (since your hmmer result sequence is "included" in the database sequence):

from Bio import SeqIO
import re

hmmer_seq_string = str(hmmer_result.seq)
dbase_seq_string = str(dbase_record.seq)
match = re.search(hmmer_seq_string, dbase_seq_string)
s = match.start()
e = match.end()
first_residues = dbase_seq_string[:s]
last_residues = dbase_seq_string[e:]
print first_residues
print last_residues