There is aready my proposal for Capablanca Random Chess. I repeat it here:

CAPABLANCA RANDOM CHESS (2004-Nov-26) Proposal 

This definition of CRC should cover the following goals:

a) creating an interesting drosophila for chess programmers 
b) using Capablancas 10x8 Chess board geometry 
c) using Capablancas piece set (incl. archbishop and chancellor) 
d) applying rules aligned to Fischer Random Chess 
e) avoiding conflicts to any claimed patents 

The CRC rules are: 

a) creating a starting position (one of 48.000): 
 1) the bishops have to be placed upon different colored
  squares; same rule applies to the implicite bishop pieces:
  queen and archbishop (aligned to FRC)
 2) the king always has to be placed somewhere between the 
  rooks to enable castlings (aligned to FRC)
 3) use only such positions without unprotected pawns (Chess)

b) describing a method of generating starting positions on 
   free squares by using a dice or random number generator: 
 1) select queen or the archbishop to be placed first (2x)
 2) place the selected 1st piece upon a bright square (5x)
 3) place the selected 2nd piece upon a dark square (5x)
 4) one bishop has to be placed upon a bright square (4x) 
 5) one bishop has to be placed upon a dark square (4x) 
 6) one chancellor has to be placed upon a free square (6x) 
 7) one knight has to be placed upon a free square (5x) 
 8) one knight has to be placed upon a free square (4x)/2 
 9) set the king upon the center of three free squares left
11) set the rooks upon the both last free squares left 
12) this establishes White's first row, the Black side 
    has to be built up symmetrically to this 
13) place ten pawns similar to traditional chess in a row 
14) skip this position if it has unprotected pawns or not
    at least three positions in line 1 differently filled
    compared to Gothic Chess (patented), this finally gives
    about 21.259 distinct starting arrays.
   
c) nature of (asymmetric Fischer-) castlings:
 1) castlings are (like in traditional chess) only valid
  if neither the affected king or rook has been moved, or
  there would be a need to jump over any third piece, or
  the king would be in chess somewhere from his starting
  position to his target field (both included). Therefore
  all squares between king and its target square (included) 
  have to be free from third pieces, same applies to the
  way the rook has to go to its target square.
 2) the alpha-castling (O-O-O, White's left side):
  like in FRC the king will be placed two rows distant
  from the border (here c-file) and the rook at the next 
  inner neighboured square.
 3) the omega-castling (O-O, White's right side):
  like in FRC the king will be placed one row distant
  from the border (here i-file) and the rook at the next 
  inner neighboured square.

d) performing castlings:
 within a GUI try to move the king upon the related rook
 or at least two squares into that direction; manually:
 1) move the king outside of the board
 2) move the rook to its end position (if need to)
 3) move the king to his end position

e) extended FEN encoding:
 1) the extended FRC-FEN could be used as a base
 2) 'a'/'A' are used to identify archbishops
 3) 'c'/'C' are used to identify chancellors
 4) '9' is used to mark nine empty squares
 5) '10' is used to encode ten empty squares
 6) if a castling enabled rook is not the most outer one
  at that side, the letter of his file has to be placed
  immediately following his castling marker symbol, where
  'q'/'Q' are used for the alpha-, 'k'/'K' for omega-side.
 
f) engine notation rules for castling moves:
 According to UCI convention the castling moves should be
 written by using both coordinates (source and target field)
 of the involved king. But there are castlings, where the 
 king does only one or none simple step. In that cases the 
 castling should be distinguishable by appending a 'k', like
 already practized in promotion moves to make them unique.
 Overmore an engine should accept O-O or O-O-O (no zeroes),
 but only use them, when the GUI would demand for such a
 less precise notation.