In general, if there are $n$ ordered observations, and a confidence level $c\%$, the $c\%$ VaR is given by the $[(1 - c\%)n + 1]^\text{th}$ highest observation. This is the observation that separates the tail from the body of the distribution. Given 260 observations, we are therefore interested in the $[(1 - 0.95) \cdot 260 + 1]^\text{th} = 14^\text{th}$ worst observation

Before the latest 10% loss, the ordered losses are as follows:
$[-9.5, -8, -7.6, -7.4, -7.2, -7.18, -7.1, -6.9, -6.57, -6.56, -6.45, -6.4, -6.25, -6.05, -5.99, -5.85]$

The 14th worst observation is -6.05. However, given that the stock decreased by 10% between yesterday and today, the arrangement changes; -10% will now be the worst return that the stock experienced over the last 260 business days. The 16 worst returns shall now be: $[-10, -9.5, -8, -7.6, -7.4, -7.2, -7.18, -7.1, -6.9, -6.57, -6.56, -6.45, -6.4, -6.25, -6.05, -5.99]$

The VaR at 95% will be based on -6.25%, which is now the 14th worst return.

$$\text{VaR} = 90 \times 6.25\% = 5.625$$

Notes.
(I) 90 here is the price of the stock today (after a 10% decline yesterday)
(II) We're working with the worst returns recorded over a 260-day moving window, so we're essentially looking back and ordering the worst returns recorded during that period, and one observation slips out of the window every day. The latest return to slip out is -3%, but this is considerably higher (less negative) than -5.85% (lowest worst return on our list), so it does not affect the makeup of the worst 16 returns.