Overlap matrix disagreement (basis set convention)

[daniel01-cpu]

Dear XTB team,

I have a doubt about the calculation of the overlap matrix in XTB. For this simple calculation in XTB:

input.xyz

2

H    0.00000000000000E+00  0.00000000000000E+00  0.00000000000000E+00
H       1.000000000      0.8      1.2
$chrg 0
$spin 0
$write
fod=true
mos=true
$end

The overlap and coefficient matrix were dumped by adding a couple lines to properties.F90 and the basis was obtained from the molden file generated by XTB

This is the basis

           1 0
 s           3   1.00000000000000
   3.37022769753360       0.273591105889083
  0.613891182214980       0.264605406538605
  0.166142911484160       8.246594753408230E-002

and this is the overlap

overlap matrix

                      1             2
     1    1.00000000 0.17976632
     2    0.17976632 1.00000000

The same calculation in pyscf gives a different overlap matrix, which agrees with orca

This is the pyscf input

import pyscf
mol = pyscf.gto.Mole()
mol.atom = 'H    0.00000000000000E+00  0.00000000000000E+00  0.00000000000000E+00; H       1.000000000      0.8      1.2'
mol.basis = {'H': pyscf.gto.basis.parse('''
 H s
   3.37022769753360       0.273591105889083
  0.613891182214980       0.264605406538605
  0.166142911484160       8.246594753408230E-002
''')}
mol.spin = 0
mol.build()
overlap = mol.intor('int1e_ovlp')
print(overlap)

and this is the overlap matrix

[[1.         0.05351956]
 [0.05351956 1.        ]]

The non-diagonal term is different (0.05351956 vs. 0.17976632)

So I guess there could be some different convention for the basis set printed in molden which makes this difference. Some orientation about this issue is greatly welcomed.

Thanks
Best
Daniel