diff --git a/scripts/release_gen.py b/scripts/release_gen.py
new file mode 100755
index 0000000000..e56a687e34
--- /dev/null
+++ b/scripts/release_gen.py
@@ -0,0 +1,418 @@
+#!/usr/bin/env python3
+import argparse
+import json
+import os
+import re
+import shlex
+import subprocess
+import sys
+import time
+from dataclasses import dataclass
+from pathlib import Path
+from typing import List, Optional, Sequence, Tuple
+
+
+def header(msg: str) -> None:
+    print(f"==> {msg}", file=sys.stderr)
+
+
+USAGE_TEXT = """
+Usage: scripts/release_gen.py [--dump-only] [-q|--quiet] [owner/repo] <from_tag> <to_tag> [version]
+
+Examples:
+  scripts/release_gen.py openai/codex v0.23.0 v0.24.0
+  scripts/release_gen.py v0.23.0 v0.24.0                # auto-detect repo from git remote
+  scripts/release_gen.py v0.23.0 v0.24.0 0.24.0         # auto-detect with explicit version
+  scripts/release_gen.py --dump-only v0.23.0 v0.24.0    # only generate releases/release_dump_<ver>.txt
+  scripts/release_gen.py -q v0.23.0 v0.24.0             # quiet Codex call with progress dots
+
+Notes:
+  - Requires: gh (GitHub CLI) for dump generation; codex CLI for note generation.
+  - If release_dump_<ver>.txt does not exist, it will be created automatically.
+  - Then runs codex to generate <ver>.txt based on the dump (unless --dump-only).
+  - If you omit tags, the script lists the last 20 releases for the repo.
+""".strip()
+
+
+# -------- argument parsing (shell-like) --------
+
+
+@dataclass
+class Args:
+    dump_only: bool
+    quiet: bool
+    rest: List[str]
+
+
+def parse_args(argv: Sequence[str]) -> Args:
+    dump_only = False
+    quiet = False
+    rest: List[str] = []
+    it = iter(argv)
+    for a in it:
+        if a == "--dump-only":
+            dump_only = True
+        elif a in ("-q", "--quiet"):
+            quiet = True
+        elif a in ("-h", "--help"):
+            print(USAGE_TEXT)
+            sys.exit(0)
+        else:
+            rest.append(a)
+    return Args(dump_only=dump_only, quiet=quiet, rest=rest)
+
+
+# -------- helpers --------
+
+
+def run(cmd: Sequence[str], check: bool = True, capture: bool = True, text: bool = True, env: Optional[dict] = None) -> subprocess.CompletedProcess:
+    return subprocess.run(cmd, check=check, capture_output=capture, text=text, env=env)
+
+
+def which(name: str) -> Optional[str]:
+    from shutil import which as _which
+
+    return _which(name)
+
+
+def abspath(p: Path) -> Path:
+    return p.resolve()
+
+
+def detect_repo_from_git() -> Optional[str]:
+    # Try origin then upstream
+    urls: List[str] = []
+    for remote in ("origin", "upstream"):
+        try:
+            cp = run(["git", "remote", "get-url", remote])
+            if cp.stdout.strip():
+                urls.append(cp.stdout.strip())
+                break
+        except subprocess.CalledProcessError:
+            continue
+    if not urls:
+        return None
+    remote = urls[0]
+    path = remote
+    # strip protocols and user@
+    for prefix in ("git@", "ssh://", "https://", "http://"):
+        if path.startswith(prefix):
+            path = path[len(prefix) :]
+    if "@" in path:
+        path = path.split("@", 1)[1]
+    # handle github.com:owner/repo or .../owner/repo
+    if ":" in path and path.split(":", 1)[0].endswith("github.com"):
+        path = path.split(":", 1)[1]
+    if "github.com/" in path:
+        path = path.split("github.com/", 1)[1]
+    path = path.lstrip("/")
+    if path.endswith(".git"):
+        path = path[: -len(".git")]
+    parts = path.split("/")
+    if len(parts) >= 2:
+        return f"{parts[0]}/{parts[1]}"
+    return None
+
+
+def show_recent_releases_and_exit(repo: str) -> None:
+    print("", file=sys.stderr)
+    print("Please pass a source/target release.", file=sys.stderr)
+    print("", file=sys.stderr)
+    print("e.g.: ./scripts/release_gen.sh rust-v0.23.0 rust-v0.24.0", file=sys.stderr)
+    print("", file=sys.stderr)
+    header(f"Recent releases for {repo}:")
+    print("", file=sys.stderr)
+    try:
+        cp = run(["gh", "release", "list", "--repo", repo, "--limit", "20"])
+        lines = cp.stdout.splitlines()
+        for line in lines:
+            if not line.strip():
+                continue
+            first = line.split()[0]
+            print(f"- {first}", file=sys.stderr)
+    except subprocess.CalledProcessError:
+        print(f"Error: unable to fetch releases for {repo}", file=sys.stderr)
+        sys.exit(1)
+    sys.exit(1)
+
+
+# -------- dump generation (ported) --------
+
+
+def gh_json(args: Sequence[str]) -> dict:
+    cp = run(["gh", *args])
+    return json.loads(cp.stdout)
+
+
+def get_tag_datetime_iso(repo: str, tag: str) -> str:
+    # Try release publish date
+    try:
+        cp = run([
+            "gh",
+            "release",
+            "view",
+            tag,
+            "--repo",
+            repo,
+            "--json",
+            "publishedAt",
+            "--jq",
+            ".publishedAt",
+        ])
+        ts = cp.stdout.strip()
+        if ts and ts != "null":
+            return ts
+    except subprocess.CalledProcessError:
+        pass
+
+    # Fallback via tag -> commit -> committer.date
+    ref = gh_json(["api", f"repos/{repo}/git/ref/tags/{tag}"])
+    obj_type = ref.get("object", {}).get("type")
+    obj_url = ref.get("object", {}).get("url")
+    commit_sha = None
+    if obj_type == "tag" and obj_url:
+        tag_obj = gh_json(["api", obj_url])
+        commit_sha = (tag_obj.get("object") or {}).get("sha")
+    else:
+        commit_sha = (ref.get("object") or {}).get("sha")
+    if not commit_sha:
+        raise RuntimeError(f"Failed to resolve commit for tag {tag}")
+    commit = gh_json(["api", f"repos/{repo}/commits/{commit_sha}"])
+    return ((commit.get("commit") or {}).get("committer") or {}).get("date") or ""
+
+
+def collect_prs_within_range(repo: str, from_iso: str, to_iso: str) -> List[dict]:
+    cp = run(
+        [
+            "gh",
+            "pr",
+            "list",
+            "--repo",
+            repo,
+            "--state",
+            "merged",
+            "--limit",
+            "1000",
+            "--json",
+            "number,title,mergedAt,author,body",
+        ]
+    )
+    data = json.loads(cp.stdout)
+
+    def keep(pr: dict) -> bool:
+        ma = pr.get("mergedAt")
+        return bool(ma and from_iso <= ma <= to_iso)
+
+    out = []
+    for pr in data:
+        if not keep(pr):
+            continue
+        out.append(
+            {
+                "number": pr.get("number"),
+                "title": pr.get("title") or "",
+                "merged_at": pr.get("mergedAt") or "",
+                "author": ((pr.get("author") or {}).get("login")) or "-",
+                "body": pr.get("body") or "",
+            }
+        )
+    out.sort(key=lambda x: x.get("merged_at") or "", reverse=True)
+    return out
+
+
+_ISSUE_CLOSING_RE = re.compile(
+    r"(?i)(?:close|closed|closes|fix|fixed|fixes|resolve|resolved|resolves)[\s:]+(?:[A-Za-z0-9_.-]+/[A-Za-z0-9_.-]+)?#(\d+)"
+)
+
+
+def format_related_issues(body: str) -> str:
+    body = body.replace("\r", "")
+    nums = {_ for _ in _ISSUE_CLOSING_RE.findall(body)}
+    if not nums:
+        return "-"
+    ints = sorted({int(n) for n in nums})
+    return ", ".join([f"#{n}" for n in ints])
+
+
+def generate_dump(repo: str, from_tag: str, to_tag: str, out_file: Path) -> None:
+    if not which("gh"):
+        print("Error: gh (GitHub CLI) is required", file=sys.stderr)
+        sys.exit(1)
+
+    header(f"Resolving tag dates ({from_tag} -> {to_tag})")
+    from_iso = get_tag_datetime_iso(repo, from_tag)
+    to_iso = get_tag_datetime_iso(repo, to_tag)
+    if not (from_iso and to_iso):
+        print(
+            f"Error: failed to resolve tag dates. from={from_tag} ({from_iso}) to={to_tag} ({to_iso})",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+
+    header("Collecting merged PRs via gh pr list")
+    prs = collect_prs_within_range(repo, from_iso, to_iso)
+    count = len(prs)
+
+    header(f"Writing {out_file} (Total PRs: {count})")
+    out_file.parent.mkdir(parents=True, exist_ok=True)
+    with out_file.open("w", encoding="utf-8") as f:
+        f.write(f"Repository: {repo}\n")
+        f.write(f"Range: {from_tag} ({from_iso}) -> {to_tag} ({to_iso})\n")
+        f.write(f"Generated: {time.strftime('%Y-%m-%dT%H:%M:%SZ', time.gmtime())}\n")
+        f.write(f"Total PRs: {count}\n\n")
+
+        for pr in prs:
+            title = pr["title"]
+            number = pr["number"]
+            merged_at = pr["merged_at"]
+            author = pr["author"]
+            body = pr["body"]
+            issues = format_related_issues(body)
+
+            f.write(f"PR #{number}: {title}\n")
+            f.write(f"Merged: {merged_at} | Author: {author}\n")
+            f.write(f"Related issues: {issues}\n\n")
+
+            dep_names = ("app/dependabot", "dependabot[bot]")
+            if author not in dep_names and re.search(r"[Dd]ependabot", author) is None:
+                f.write("Description:\n")
+                max_len = 2000
+                snippet = body[:max_len]
+                if len(body) > max_len:
+                    snippet += "..."
+                f.write(snippet + "\n\n")
+            f.write("-----\n\n")
+
+    header(f"Done -> {out_file}")
+
+
+def build_prompt(dump_path: Path, gen_file: Path) -> str:
+    dump_content = dump_path.read_text(encoding="utf-8")
+    example = (
+        "## Highlights:\n\n"
+        "### New Features\n"
+        "- Queued messages (#2637)\n"
+        "- Copy Paste / Drag & Drop image files (#2567)\n"
+        "- Add web_search tool (#2371)\n"
+        "- Add transcript mode (Ctrl+T) with scrolling ability (#2525)\n"
+        "- Edit/resume conversation (esc-esc) from previous messages (#2607)\n\n"
+        "### TUI\n"
+        "- Hide CoT by default; show headers in status indicator (#2316)\n"
+        "- Show diff output in pager (+ with hunk headers) (#2568)\n"
+        "- Simplify command approval UI (#2708)\n"
+        "- Unify Esc/Ctrl+C interrupt handling (#2661)\n"
+        "- Fix windows powershell paste (#2544)\n\n"
+        "### Tools and execution\n"
+        "- Add support for long-running shell commands `exec_command`/`write_stdin` (#2574)\n"
+        "- Improve apply_patch reliability (#2646)\n"
+        "- Cap retry counts (#2701)\n"
+        "- Sort MCP tools deterministically (#2611)\n\n"
+        "### Misc\n"
+        "- Add model_verbosity config for GPT-5 (#2108)\n"
+        "- Read all AGENTS.md files up to git root (#2532)\n"
+        "- Fix git root resolution in worktrees (#2585)\n"
+        "- Improve error messages & handling (#2695, #2587, #2640, #2540)\n\n\n"
+        "## Full list of merged PRs:\n\n"
+        " - #2708 [feat] Simplfy command approval UI\n"
+        " - #2706 [chore] Tweak..."
+    )
+    instr = (
+        f"{dump_content}\n\n---\n\n"
+        f"Please generate a summarized release note based on the list of PRs above. "
+        f"Then, write a file called {gen_file} with your suggested release notes. "
+        f"It should follow this structure (+ the style/tone/brevity in this example):\n\n"
+        f"\"{example}\""
+    )
+    return instr
+
+
+def run_codex(prompt: str, quiet: bool) -> int:
+    if not which("codex"):
+        print(
+            "Error: codex CLI is required for generation. Use --dump-only to skip.",
+            file=sys.stderr,
+        )
+        return 127
+
+    cmd = ["codex", "exec", "--sandbox", "workspace-write", prompt]
+    if quiet:
+        try:
+            proc = subprocess.Popen(
+                cmd,
+                stdout=subprocess.DEVNULL,
+                stderr=subprocess.DEVNULL,
+                text=True,
+            )
+        except FileNotFoundError:
+            return 127
+        while True:
+            ret = proc.poll()
+            if ret is not None:
+                break
+            print(".", end="", file=sys.stderr, flush=True)
+            time.sleep(1)
+        print("", file=sys.stderr)
+        return ret or 0
+    else:
+        # stream to stderr like the shell script
+        proc = subprocess.run(cmd, stdout=sys.stderr, stderr=sys.stderr, text=True)
+        return proc.returncode
+
+
+def main(argv: Sequence[str]) -> int:
+    pargs = parse_args(argv)
+
+    rest = pargs.rest
+    # repo optional first arg
+    if rest and "/" in rest[0]:
+        repo = rest[0]
+        rest = rest[1:]
+    else:
+        repo = detect_repo_from_git() or ""
+        if not repo:
+            print(
+                "Error: failed to auto-detect repository from git remote. Provide [owner/repo] explicitly.",
+                file=sys.stderr,
+            )
+            return 1
+
+    if len(rest) < 2:
+        show_recent_releases_and_exit(repo)
+        return 1  # unreachable
+
+    from_tag, to_tag = rest[0], rest[1]
+    ver = rest[2] if len(rest) >= 3 else to_tag
+    ver = ver.lstrip("v")
+
+    script_dir = Path(__file__).resolve().parent
+    releases_dir = script_dir / "releases"
+    dump_file = releases_dir / f"release_dump_{ver}.txt"
+    gen_file = releases_dir / f"{ver}.txt"
+
+    # Create dump if missing
+    if not dump_file.exists():
+        header(f"Dump not found: {dump_file}. Generating...")
+        generate_dump(repo, from_tag, to_tag, dump_file)
+    else:
+        header(f"Using existing dump: {dump_file}")
+
+    if pargs.dump_only:
+        return 0
+
+    dump_path = abspath(dump_file)
+    prompt = build_prompt(dump_path, gen_file)
+    header(f"Calling codex to generate {gen_file}")
+    status = run_codex(prompt, pargs.quiet)
+
+    if gen_file.exists():
+        # Output only the generated release notes to stdout
+        sys.stdout.write(gen_file.read_text(encoding="utf-8"))
+        return 0
+    else:
+        print(f"Warning: {gen_file} not created. Check codex output.", file=sys.stderr)
+        return 1 if status != 0 else 1
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv[1:]))
+