The way ChatItem used to work was by sending the client the data of the item as a string. To achieve that, the plugin used reflection to get the item's NBT Tag and call the toString() method on it. However, that method did not escape special characters or ignore invalid NBT Tags. Therefore, tag keys containing values with special JSON meaning, like '['s, ']'s or '{'s, '}'s, ','s and ':'s, would fail to be properly parsed by the client, therefore the client resorting to drawing the "Invalid Item!" string instead of the proper item representation.
In this new version, I have created a new method to replace the NBTTag#toString() that ignores invalid tags and other tags specified in the list of ignored tags, and which properly escapes any special characters.

As some of you might have already realized, the plugin's original idea is that instead of editing a chat event, it intercepts outgoing chat packets from the server. This means it will work for ANY message the client receives. However, those messages sent in packets are wrapped in JSON. My plugin's core consists of a class whose sole purpose is to parse the JSON strings of the packet into another JSON string which replaced existing occurences of placeholders with their replacement. However, it look like I only made it to search the JSON String in a "shallow" way. Here's an example to show you what I mean:
While this would be parsed properly:
http://i.imgur.com/3dTTDBk.png
This wouldn't!
http://i.imgur.com/DltcGaB.png

All of this because my original code would only look for a text element inside each object of the JsonArray, while, as you can see, the other objects contains no text field, rather another "extra" array with a text object inside it. This time, I modified my code to recursively search the entire JSON string. It will take a bit more time for the parser to work, but that shouldn't be an issue anyway as it will only postpone the chat for about 1-2 ms, and as it's running in an async thread, it should have no impact on actual server performance.