If you are planning to do a pairwise alignment, you need to have a single file for each query and target fasta sequence, kind of like this. You could read the existing files using your favorite fasta parsing modules in either Bio-Perl or Bio-Python and write each sequence into a new file.

>Query1
TGCCTACTGAGCTGAAACAGT
>Target1
CAGTAACCATGACCTCCCGCAGGACAGCGGAGCC

Here's a thread on splitting fasta files: How To Split A Multiple Fasta

all sequences must be in 1 file, one after another

This means just keep all the fasta sequences in one file, nothing more. When clustalW asks for the file name containing fasta sequences give the file name in which all the fasta sequences are present. You can set all other parameters like MSA or pairwise alignment parameters before the alignment begins.

The full quote is:

SEQUENCE INPUT: all sequences must be in 1 file, one after another.
7 formats are automatically recognised: NBRF-PIR, EMBL-SWISSPROT,
Pearson (Fasta), Clustal (*.aln), GCG-MSF (Pileup), GCG9-RSF and GDE flat file.
All non-alphabetic characters (spaces, digits, punctuation marks) are ignored
except "-" which is used to indicate a GAP ("." in MSF-RSF).

This gives you a list of possible sequence formats. Most of them are well documented, in particular the FASTA format. You can also find more examples in the EMBOSS documentation.

This said, as suggested by the other answers, the FASTA format may be the easiest for you.