If some parameter with asymmetric error bars, it will be interpreted as a distribution that pastes two half gaussians together, with their sigmas being plus and minus error bars.
However, this does not conserve the coverage, meaning that the coverage will no longer be 68%.