Constant oil price as a forecast with GARCH is unlikely. Check for errors and try different GARCH parameters: p, q.

Reference: Article Modeling and Forecasting the Crude Oil Price in Nigeria

This doesn't sound right, there should be a mistake in your parameters.

Before attempting to address your real issue, let me point out that it is not wise to forecast for an entire year. Here, you are looking forward to forecast for nearly 250 data points (taking 5 trading days a week). This is too long a period to forecast for, especially for crude oil market which we know to be highly volatile. Moreover, the longer the prediction period the lesser will be the accuracy. Put differently, the realized values tend to deviate more from the predicted value for distant periods. Thus, for a reasonably accurate prediction, I suggest that you keep the prediction period small - say a month max.

Coming to your problem, as obvious, there is something that needs to be corrected in your code. What I feel is that you might have designed your code such that - for predicting the value for 04.01.2019, you continue to use the same historical data from 02.01.1986 to 02.01.2019 WITHOUT including the newly computed value for 03.01.2019. Similarly, for estimating for 05.01.2019, you need to include the estimates of both 03.01.2019 and 04.01.2019 along with the actual observations and so on.

Lastly, the oil price dynamics change very drastically every decade, if not every few years. This has to do with the increasing financialisation of crude market coupled with geo-political and strategic interventions. Thus, in your attempt to forecast crude prices, including data pertaining to 1980's and 1990’s might affect your forecast results adversely. Data from 2010 to 2018 should be good enough to make a reasonable accurate prediction. Just my thoughts... You may think it through. Hope it helps.