diff --git a/CLAUDE.md b/CLAUDE.md
index 7408dc4..992dead 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -7,8 +7,13 @@
     - `./references.xml`: Contains reference data which is linked mostly in the metadata, for locations, people and definitions for terms used in descibing letter heritage data.
     - `./traditions.xml`: Contains data about the traditions of the letters, such as the edition they were published in, and the publisher, as well as - often times - additional information, other versions (or drafts) of the letters, and so on.
     - `./edits.xml`: Contains data about the editorial changes made to letters - often times comparing the edited text of sources to the original and restoring the original state as well as possible -, cross referencing edit-elements in the letter fulltext.
-    - `./marginals.xml`: Contains commentary on letters, referencing them by ID, page- and line-numbers.
+    - `./Marginal-Kommentar.xml`: Contains commentary on letters, referencing them by ID, page- and line-numbers.
+    - `./Bibel-Kommentar.xml`, `forschung.xml`, `Register-Kommentar.xml`: Contains a glossary of referenced bible verses, a list of research titles, and a register of terms, but mostly persons and works referenced and linked in the letter commentary.
 
 ## Rules
-- Rules for XML transformations:
-    -
+- What to do when an XML transformation is requested:
+    - Before transforming the XML, give a few examples of how you understand the users request and how you would transform the XML. TRy to use a diverse set of cases.
+    - After giving examples, you'll likely will need to write a python script to transform the XML. Use the `lxml.etree` library to parse the XML in python and transform it. If text should be transformed, try to find the right place in XML witth said library, then you are allowed to use regex to transform text.
+    - Before running the script, make sure to test it on a small subset of the XML data, so you can see if it works as expected. Alternatively, generate test data to test the transformations against. You are only allowed to do transformations the user asked for, and nothing else.
+    - After transformation and testing, you may run a formatter in python also using `lxml.etree` to format the XML output, so it is readable and well structured.
+    - Save all generated scripts in the `scripts/` directory, so they can be reused later. Overwrite the ortiginal xml files only if the transformation was deemed sucessful by the user.