I'm confused about the exact multiplicative factor that relates Goldman symplectic form on the $\operatorname{SL}(2,\mathbb R)$-character variety and the Weil–Petersson symplectic form on Teichmüller space. 

One way to define Goldman symplectic form is the following:
$$
    \omega_{h} = \int_{S} \operatorname{trace} (\dot{\nabla} \wedge \smash{\dot{\nabla}}')
$$
where $\dot{\nabla}$ and $\smash{\dot{\nabla}}'$ are variations of the Levi-Civita connection of the hyperbolic metric $h$. If those variations are induced by divergence-free, $h$-self-adjoint, traceless endomorphisms $\dot{J}$ and $\dot{J}'$, then it is easy to see that
$$
      \dot{\nabla} = -\frac{1}{2} J\nabla\dot{J} \ .
$$
Now, Goldman claims that this form is $-8$ times the Weil-Petersson symplectic form, but, integrating by parts, I get instead a multiplicative factor of $+4$. What am I doing wrong?