I am a novice at DSGE. Now I have an uncertainty: is the initial shock value of impulse response a standard deviation of shock?
Linear models are scale-invariant, so the IRF scale linearly. By default, Dynare will show IRFs to a one standard deviation shock.