While computing the mechanism of a trimethylsilylcyanation related to this work (10.1021/acs.organomet.1c00070) I managed to optimize the geometry of both reactants and product and locate a good TS (vibration of the imaginary frequency fits with the breaking/formation of the new bond).

However the IRC path shows some higher energy strucutres around the TS (red dot in the figure) corresponding to stretched CN bond.

In addition, while the IRC connects to the product in the forward direction, it does not connect to the reactants in the backward direction (the maximum number of steps is not reached). Thinking of an intermediate, I also optimize the geometry from this IRC calculation, but it ultimately goes to the reactants...

I would be grateful if you can help me sorting this IRC calculation that doesn't seems correct.