diff --git a/.github/guid-index.json b/.github/guid-index.json
index bf081cf5..ef0bda2d 100644
--- a/.github/guid-index.json
+++ b/.github/guid-index.json
@@ -3075,7 +3075,7 @@
     "path": "scripts/assemble_todo.py",
     "guid": "af7ef324-6c69-411e-b1a9-98c9ba2b31e3",
     "title": "assemble todo",
-    "version": "1.1.0",
+    "version": "1.2.0",
     "header_path": "scripts/assemble_todo.py",
     "path_mismatch": false
   },
diff --git a/changelog.d/20260719-assembler-width-stable.md b/changelog.d/20260719-assembler-width-stable.md
new file mode 100644
index 00000000..4180d9b6
--- /dev/null
+++ b/changelog.d/20260719-assembler-width-stable.md
@@ -0,0 +1,15 @@
+### Fixed
+
+#### `assemble_todo.py` is now formatted identically at any ruff line-length
+
+`ruff format` is not idempotent across line-lengths: a file formatted at 80
+columns gets its short lines rejoined at 88. The assembler is copied verbatim
+into repos that have no `ruff.toml` of their own, where ruff falls back to its
+88-column default — so `ruff format --check` failed there even though the file
+was correctly formatted for this repo.
+
+Every multi-line construct now carries a magic trailing comma (which pins it
+exploded at any width), and the two constructs that could not take one were
+reshaped to fit under 79 columns. Verified byte-identical output at widths 79,
+80, 88, 100, 120 and 200, and the file passes `ruff check`/`ruff format --check`
+under both this repo's config and a bare `--isolated` default.
diff --git a/scripts/assemble_todo.py b/scripts/assemble_todo.py
index ca48fd40..edfc7d17 100755
--- a/scripts/assemble_todo.py
+++ b/scripts/assemble_todo.py
@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
 # file: scripts/assemble_todo.py
-# version: 1.1.0
+# version: 1.2.0
 # guid: af7ef324-6c69-411e-b1a9-98c9ba2b31e3
 # last-edited: 2026-07-19
 
@@ -50,10 +50,12 @@
 HTML_COMMENT = re.compile(r"<!--.*?-->", re.DOTALL)
 
 VERSION_HEADER = re.compile(
-    r"^(<!--\s*version:\s*)(\d+)\.(\d+)\.(\d+)(\s*-->)$", re.M
+    r"^(<!--\s*version:\s*)(\d+)\.(\d+)\.(\d+)(\s*-->)$",
+    re.M,
 )
 LAST_EDITED_HEADER = re.compile(
-    r"^(<!--\s*last-edited:\s*)(\S+)(\s*-->)$", re.M
+    r"^(<!--\s*last-edited:\s*)(\S+)(\s*-->)$",
+    re.M,
 )
 
 
@@ -81,9 +83,8 @@ def find_fragments(fragment_dir: Path) -> list[Path]:
     """
     if not fragment_dir.is_dir():
         return []
-    return sorted(
-        path for path in fragment_dir.glob("*.md") if path.name != "README.md"
-    )
+    candidates = fragment_dir.glob("*.md")
+    return sorted(p for p in candidates if p.name != "README.md")
 
 
 def fragment_body(path: Path) -> str:
@@ -121,7 +122,9 @@ def _bump(match: re.Match[str]) -> str:
         print("warning: no '<!-- version: X.Y.Z -->' header to bump.")
 
     text, count = LAST_EDITED_HEADER.subn(
-        lambda m: f"{m.group(1)}{today}{m.group(3)}", text, count=1
+        lambda m: f"{m.group(1)}{today}{m.group(3)}",
+        text,
+        count=1,
     )
     if not count:
         print("warning: no '<!-- last-edited: ... -->' header to refresh.")
@@ -136,12 +139,13 @@ def insert_at_marker(text: str, marker: str, addition: str) -> str:
     scriv uses for release sections.
     """
     marker_line = re.compile(
-        rf"^([ \t]*<!--\s*{re.escape(marker)}\s*-->[ \t]*)$", re.M
+        rf"^([ \t]*<!--\s*{re.escape(marker)}\s*-->[ \t]*)$",
+        re.M,
     )
     match = marker_line.search(text)
     if not match:
         raise AssemblyError(
-            f"Output file has no '<!-- {marker} -->' marker; refusing to guess where tasks belong."
+            f"No '<!-- {marker} -->' marker in the output file; refusing to guess where tasks belong.",
         )
     return text[: match.end()] + f"\n\n{addition}" + text[match.end() :]