I think it might be a bug, but given the very few resources currently given to ECCE, it's better to develop work-arounds than to demand quick bug fixes.
A reasonable criticism of this post would be that it really doesn't matter in the long run though, as the bonds that we make here are eye-candy only -- DFT doesn't 'know' what bonds are in the sense of the ball-and-stick model. On the other hand, ECCE will complain loudly if there's a discrepancy between oxidation states, multiplicity and number of vacant sites.
In the end, if you have a crystal structure to start from, use that. If not, then this is a fairly simple way of yielding a visually pleasing and reasonably intuitive model.
Anyway, let's use Noyori's catalyst as an example:
Start ECCE, set up a new job and click on Builder.
Click on Import from Structure Library, select Alicycles, and click on Cyclooctane, then click in the main window.
Delete one hydrogen on each of the sp2 carbons, and change the bond type to double bond:
Change the sp2 carbons to TrigonalPlanar, and Clean (click the broom which does MM (UFF) geometry optimization).
Insert an octahedral iridium atom -- you'll need two 'nubs' for each pi bond, and two nubs for the two phosphine ligands.
Change the sp2 carbons back to tetrahedral so you get 'nubs' to bond to (not supposed to be necessary according to the manual, but it is). Make Metal-pi bonds.
Insert two tetrahedral phosphorous atoms, and make single bonds
At this point the structure isn't quite perfectly square planar like we would like it to be. The easiest way to sort that out is to set the torsional angles between the H-C(=)-Ir-P to 0 (uncheck rotate group around bond) for each P, then Clean. The result isn't perfect, but still not too shabby: