diff --git a/scripts/release-changelog.py b/scripts/release-changelog.py
new file mode 100755
index 000000000..5e15b7254
--- /dev/null
+++ b/scripts/release-changelog.py
@@ -0,0 +1,96 @@
+#!/usr/bin/env python3
+"""Generate release notes from CHANGELOG.md for use in GitHub Releases (or anywhere).
+
+Usage::
+
+    python scripts/release-changelog.py VERSION
+    python scripts/release-changelog.py 1.1.0
+    python scripts/release-changelog.py --file path/to/CHANGELOG.md 1.1.0
+
+Extracts the requested ``## [VERSION]`` section (heading and ``### …``
+subheadings included) and prints it to stdout. The only transformation
+applied is collapsing each entry paragraph onto a single line, so it is
+suitable for pasting into release notes that don't need 80-column wrapping.
+``(PR [...])`` references stay on their own two-space-indented continuation
+line. The input file is never modified.
+
+Every paragraph is unfilled. Indentation is preserved — each logical line
+(bullets, sub-bullets, and the ``(PR [...])`` continuation) keeps its
+leading whitespace; only the wrapped continuation lines that follow them
+get joined back. Code-block paragraphs (triple-backtick fences, or deeply
+indented blocks with no list markers around) are passed through untouched.
+"""
+
+from __future__ import annotations
+
+import argparse
+import re
+import sys
+from pathlib import Path
+
+DEFAULT_CHANGELOG = Path(__file__).resolve().parent.parent / "CHANGELOG.md"
+
+PARAGRAPH_SPLIT_RE = re.compile(r"(\n[ \t]*\n+)")
+CONTINUATION_RE = re.compile(r"\n(?![ \t]*[-*+] )[ \t]+")
+PR_RE = re.compile(r" \(PR \[")
+CODE_FENCE_RE = re.compile(r"^\s*```")
+DEEP_INDENT_RE = re.compile(r"^ {4,}")
+LIST_MARKER_RE = re.compile(r"^[ \t]*[-*+] ")
+SECTION_HEADING_RE = re.compile(r"^## \[", re.MULTILINE)
+
+
+def is_code_block_paragraph(paragraph: str) -> bool:
+    lines = paragraph.splitlines()
+    if any(CODE_FENCE_RE.match(line) for line in lines):
+        return True
+    # In a list paragraph, 4-space indent is a sub-bullet continuation, not
+    # code. Only treat deep-indented lines as code when no bullets are around
+    # to claim them.
+    if any(LIST_MARKER_RE.match(line) for line in lines):
+        return False
+    return any(DEEP_INDENT_RE.match(line) for line in lines)
+
+
+def unfill_entry(paragraph: str) -> str:
+    joined = CONTINUATION_RE.sub(" ", paragraph)
+    return PR_RE.sub("\n  (PR [", joined, count=1)
+
+
+def process(text: str) -> str:
+    parts = PARAGRAPH_SPLIT_RE.split(text)
+    for i in range(0, len(parts), 2):
+        para = parts[i]
+        if para and not is_code_block_paragraph(para):
+            parts[i] = unfill_entry(para)
+    return "".join(parts)
+
+
+def extract_section(text: str, version: str) -> str:
+    head_re = re.compile(rf"^## \[{re.escape(version)}\]", re.MULTILINE)
+    m = head_re.search(text)
+    if not m:
+        sys.exit(f"error: version {version!r} not found")
+    nxt = SECTION_HEADING_RE.search(text, m.end())
+    end = nxt.start() if nxt else len(text)
+    return text[m.start() : end]
+
+
+def main() -> None:
+    summary = (__doc__ or "").splitlines()[0]
+    parser = argparse.ArgumentParser(description=summary)
+    parser.add_argument("version", help="Version to extract (e.g. 1.1.0).")
+    parser.add_argument(
+        "--file",
+        type=Path,
+        default=DEFAULT_CHANGELOG,
+        help=f"Path to the changelog (default: {DEFAULT_CHANGELOG}).",
+    )
+    args = parser.parse_args()
+
+    text = args.file.read_text()
+    section = extract_section(text, args.version)
+    sys.stdout.write(process(section))
+
+
+if __name__ == "__main__":
+    main()