The energy function says it dislikes your protein by claiming there is a diketopiperazine —does it have anything odd? Say GFP chromophore, MIO cofactor encoded oddly.
Does it load after stripping in pymol everything bar peptides? Rosetta is fine with crosslinks (LINK records in PDBs) and CONECT records are ignored mostly but this would clean up bad formatting.
What scorefunction are you using?