I would like to prove the context-free language $$ \mathcal{A} = \{ w\#x ~:~ w^R \text{ is a substring of $x$ for } w,x \in \{0,1\}^* \}, $$
has the context free grammar
\begin{align*} S &\rightarrow TX\\ T &\rightarrow 0T0 ~|~ 1T1 ~|~ \#X\\ X &\rightarrow 0X ~|~ 1X ~|~ \epsilon. \end{align*}
However, it is not clear to me how this is even true. First I examine $X$ and note $X$ generates a string with arbitrary mix of $0$'s and $1$'s. Furthermore the string could begin or end with $0$ or $1$. But I do not see how $T$ could generate a string whose reverse is a substring of $X$. In particular, $T$ is defined in terms of $X$, so isn't $T$ always longer than $X$?
As a simple example, let $X = 0$, and $T = \#X = \#0$, then $S = \#00$, and this string is not in $\mathcal{A}$!
You’re correct in thinkint that the proposed grammar doesn’t work. What you want to do is generate $w$ and $w^R$ first, much as you’ve tried to do with $T$, and then take care of the rest of $x$ and the $\#$. Assuming that you don’t want $S$ to appear on the righthand side of any production, you might begin with this:
$$\begin{align*} &S\to 0T0\mid 1T1\mid X\\ &T\to 0T0\mid 1T1\mid X \end{align*}\tag{1}$$
When you finally generate $X$, you’ll have $wXw^R$ for some $w\in\{0,1\}^*$. Clearly the $X$ can be used to generate $\#u$ for any $u\in\{0,1\}^*$, so we can extend $(1)$ with some $X$ productions to generate anything of the form $w\#uw^R$, but this isn’t quite good enough: it forces the $w^R$ to be at the righthand end of $x$.
If $w$ is the empty word, this isn’t a problem, so the production $S\to X$ is fine, but if $w$ is not the empty word, we must be sure to allow for something to the right of $w^R$. We can do this by modifying the first line of $(1)$ slightly:
$$\begin{align*} &S\to 0T0Y\mid 1T1Y\mid X\\ &T\to 0T0\mid 1T1\mid X \end{align*}\tag{2}$$
The grammar fragment in $(2)$ will give us $X$ and anything of the form $wXw^RY$, and you shouldn’t have too much trouble figuring out what $X$ and $Y$ productions you need to add to $(2)$ to get a grammar that generates $\mathcal{A}$. Once you’ve done that, we can worry about actually proving that your grammar works as intended.