How can I filter out text twice in Powershell?

The simplest PSv3+ solution is to use PowerShell's built-in XML DOM support, which makes an XML document's nodes accessible as a hierarchy of objects with dot notation:

PS> ([xml] (Get-Content -Raw txtfile.txt)).CustomObject.fields.fullName
ActivityDate
ActivityDateTime
Guid
Description    

Note how even though .fields is an array - representing all child <fields> elements of top-level element <CustomObject> - .fullName was directly applied to it and returned the values of child elements <fullName> across all array elements (<field> elements) as an array.

This ability to access a property on a collection and have it implicitly applied to the collection's elements, with the results getting collected in an array, is a generic PSv3+ feature called member enumeration.

As an alternative, consider using the Select-Xml cmdlet (available in PSv2 too), which supports XPath queries that generally allow for more complex extraction logic (though not strictly needed here); Select-Xml is a high-level wrapper around the [xml] .NET type's .SelectNodes() method.
The following is the equivalent of the solution above:

$namespaces = @{ ns="http://soap.force.com/2006/04/metadata" }
$xpathQuery = '/ns:CustomObject/ns:fields/ns:fullName'
(Select-Xml -LiteralPath txtfile.txt $xpathQuery -Namespace $namespaces).Node.InnerText

Note:

Unlike with dot notation, XML namespaces must be considered when using Select-Xml.

Given that <CustomObject> and all its descendants are in namespace xmlns, identified via URI http://soap.force.com/2006/04/metadata, you must:

• define this namespace in a hashtable you pass as the -Namespace argument
  • Caveat: Default namespace xmlns is special in that it cannot be used as the key in the hashtable; instead, choose an arbitrary key name such as ns, but be sure to use that chosen key name as the node-name prefix (see next point).
• prefix all node names in the XPath query with the namespace name followed by :; e.g., ns:CustomObject