zh locale: translate skill names and descriptions

skilldesc.txt takes Longju's name, description and condition columns
and its affect texts where their format arguments match 40250's; a
reviewed overlay fills the gaps and renames the empire languages.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

scripts/build_zh_locale.py

@@ -31,6 +31,8 @@ KEYED_FILES = ("locale_game.txt", "locale_interface.txt")
 # Image folders whose labels are drawn text (the character window's tabs and stat labels). Each is taken
 # from Longju as a whole: its .sub files carry their own coordinates into the folder's atlas.
 IMAGE_DIRS = ("ui/windows",)
+# skilldesc.txt columns taken from Longju: names, description, conditions (2-8), affect texts (17/20/23).
+SKILL_COLUMNS = (2, 3, 4, 5, 6, 7, 8, 17, 20, 23)
 
 
 def keyed_lines(path: Path, encoding: str):
@@ -71,6 +73,42 @@ def convert_names(source: Path, destination: Path, overlay: Path = None):
     return {"converted": converted, "overlaid": len(overlaid), "unrepresentable_vnums": unrepresentable}
 
 
+def merge_skilldesc(en: Path, zh: Path, destination: Path, overlay: Path = None):
+    """40250 rows with Longju's text columns; an affect text keeps the English one unless its format
+    arguments match, since the client passes it to _snprintf with the min/max values."""
+    chinese = {}
+    for line in zh.read_bytes().decode("gb18030").splitlines():
+        parts = line.split("\t")
+        if parts[0].isdigit():
+            chinese[parts[0]] = parts
+    reviewed = {}
+    if overlay and overlay.is_file():
+        for line in overlay.read_text(encoding="utf-8").splitlines()[1:]:
+            vnum, column, text = line.split("\t")
+            reviewed[(vnum, int(column))] = text
+    rows, translated, incompatible = [], 0, []
+    for line in en.read_bytes().decode("cp1252").splitlines():
+        parts = line.split("\t")
+        if parts[0].isdigit():
+            source = chinese.get(parts[0], [])
+            for column in SKILL_COLUMNS:
+                if column >= len(parts):
+                    continue
+                text = reviewed.get((parts[0], column))
+                if text is None and column < len(source) and source[column]:
+                    text = source[column]
+                if text is None:
+                    continue
+                if FORMAT.findall(parts[column]) != FORMAT.findall(text):
+                    incompatible.append(f"{parts[0]}:{column}")
+                    continue
+                parts[column] = text
+                translated += 1
+        rows.append("\t".join(parts))
+    destination.write_bytes(("\n".join(rows) + "\n").encode("gbk"))
+    return {"translated_cells": translated, "incompatible_format_cells": incompatible}
+
+
 def tool(name: str) -> Path:
     for build_dir in BUILD_DIRS:
         path = REPO / build_dir / "src/port" / name
@@ -169,6 +207,11 @@ def build(en: Path, zh: Path, output: Path, client_base: Path,
         if (zh / filename).is_file():
             shutil.copyfile(zh / filename, locale_dir / filename)
 
+    if (zh / "skilldesc.txt").is_file():
+        report["files"]["skilldesc.txt"] = merge_skilldesc(
+            en / "skilldesc.txt", zh / "skilldesc.txt", locale_dir / "skilldesc.txt",
+            names_dir / "skilldesc_zh_overlay.txt" if names_dir else None)
+
     for folder in IMAGE_DIRS:
         english_files = {path.name for path in (en / folder).iterdir()}
         chinese_files = {path.name for path in (zh / folder).iterdir()}