A "Clipping Mask" node should take a secondary input for the mask and cause it to clip. The node should have a secondary input that receives some graphical data to use as the clipping mask.

But if nothing is provided, it should instead default to clipping with the layer beneath this one in the stack. It does that by adding a "clip on layer beneath" field on the blending options for the layer, which will cause the Merge node to apply the blending as desired.

One more possibility is that we could change the blending options struct (which stores the opacity and blend mode) into a composable lambda. This would generalize opacity and blend modes, and other sorts of blending properties, into something controllable by nodes. The Merge node would then simply apply that lambda. Perhaps the lambda itself could be represented as a node graph.