This is harder than it should be at present. Your options are:
(i) use 'batch_transform' in either daily or minutely mode to accumulate OHLC or just L data. But this is soon(ish) to be deprecated and I've never felt it worked properly in minutely algos.
(ii) use 'history' in a minutely algo to give you 11 closing prices (10 days plus the current minute) if you wanted the low of the closes or 11 lows if you wanted the low of the day. But this runs every minute so you may just want to look at the opening (09:31) bar and skip the rest of the day.
(iii) use another structure like a deque to accumulate minutely L data but this might get confusing with late opening/early closing days.
(iv) find a way of using talib.MIN() but if you were collecting data with 'history' anyway this seems pointless
The attached uses 'history' at 09:31, skips the rest of the day, and records the low of the 10 days excluding the current minute.