I checked the link in below it shows LT couldn't select correct word! and it selected the last word at the sentence instead of correct word. for example at phrase
This is a foo.
The rule should select foo but it selected other word (like this)
In our case

            <rule>
                <pattern>
                    <token>دست</token>
                    <token regexp="yes">(نویس|خوش|پاچه|چین|آورد)</token>
                </pattern>
                <message>Do you mean  «<suggestion><match no="1" regexp_match="([^إأةؤورزژاآدذ])$" regexp_replace="$1‌"/>\2</suggestion>»?</message>
                <short>correcting ZWNJ</short>
                <example type="incorrect">
                    <marker>دست آورد</marker>
                </example>
                <example type="correct">دست‌آورد</example>
            </rule>

Follow up to this, main.css needs a RTL pair for RTL languages. It can be generated online based on main.css or be done once using CSSJanus

I use wiki page check to find grammar errors in my native language part of wikipedia, so if the marked text really has an obvious error I will follow edit link and make a correction at wikipedia web site. This way check a random page is a very nice feature. However, many pages are already correct from LT point of view and I need to do several clicks on "check page" button before it shows page with errors. So it would be nice if the there is a checkbox "Show only pages with errors" or this can be a default behaviour.

The online WikiCheck only allows to check pages on Wikipedia. It would be nice if it could check pages on other Wikimedia sites as well, like Wikibooks or Wikisource. They all use the Mediawiki API, so in theory there shouldn't be many changes required for this.