feat(skills): add release announcement workflow

.agents/skills/release-announcement/SKILL.md

@@ -0,0 +1,26 @@
+---
+name: release-announcement
+description: 从本项目最新的 release-notes Markdown 制作微信群发布短文案与竖版长图，核验贡献者 GitHub ID 并嵌入真实头像。用于版本发布宣传，不负责创建或发布 GitHub Release。
+---
+
+# Release announcement
+
+在本仓库调用 `$release-announcement` 时，交付一张可发微信群的 PNG 长图和一段很短的群文案。默认使用 [版式与文案要求](references/editorial-brief.md)；用户另有要求时以用户要求为准。
+
+调用方式：Codex 使用 `$release-announcement`；Claude Code Haha 的终端输入 `/release-announcement`。这两个工具都从项目的 `.agents/skills/` 读取同一份技能。
+
+## 工作流
+
+1. 在仓库根目录运行 `python3 .agents/skills/release-announcement/scripts/release_poster.py scan --repo . --out artifacts/release-posters/latest-scan.json`。它按语义版本比较工作区、当前 Git 历史、本地 `main` 和已有的 `origin/main`，不切换分支、不抓取远程。读取 JSON 中的中文发布说明、来源和 `@GitHubID` 提及；`credit_candidate` 仅表示有明确署名格式，仍需核对上下文。如果来源不是工作区文件，明确告知用户。
+   复测旧版本时加 `--version 0.6.6` 并使用独立输出路径，避免覆盖最新版本的 scan 结果。
+2. 仅依据该版本的中文 release note 提炼主要功能和关键修复；纯修复版的 `features` 留空，纯功能版的 `fixes` 留空，绝不为了填版式虚构内容。有重要模型或供应商更新时可加一个简短区块。每条信息必须能追溯到原文，压缩表述时保留条件和边界。安装提醒只写原文确有的信息。不要把英文译文重复计数。
+3. 对 `credit_candidate: true` 的账号逐个核对发布说明里的贡献项；普通 `@ID` 提及（例如 PR reviewer）不算代码署名。`render` 会通过 GitHub 公共用户 API 校验 ID 并下载真实头像；无法核验时停止并报告，不能臆造头像、错配姓名，不能让图像生成模型重绘头像。正式生成不能使用 `avatar_path` 跳过核验。
+4. 按 [数据格式](references/poster-spec.md) 写 `poster-spec.json`，版本必须与 scan 结果一致。运行 `python3 .agents/skills/release-announcement/scripts/release_poster.py render --scan artifacts/release-posters/latest-scan.json --spec <poster-spec.json> --out <poster.png>`。建议输出到 `artifacts/release-posters/v<version>/`，保留已有文件，文件名加版本或修订号。此脚本固定使用紧凑版式绘制准确中文和头像；不要用 imagegen 绘制正文。
+5. 写一段适合微信群的短文案，保存为同目录的 `wechat-copy.txt`：版本号 + 1 句主要变化 + 1 句感谢，具体细节交给图片。有本次明确署名的贡献者时感谢代码贡献；没有时只感谢反馈与支持，不暗示本版有特定群友提交。文案不要宣称未核实的下载地址或公开发布状态，也不要自动向群发送。若需要下载入口，先检查对应 GitHub Release 是否存在。
+6. 打开成图检查标题、文字换行、头像和 ID 对应、正文是否从顶部很快开始。最后提供 PNG、短文案、所用 release note 来源和任何未核实项。生成物放在被 Git 忽略的 `artifacts/`，不要提交生成图、头像缓存或中间 JSON。
+
+## 图像与环境
+
+默认长图不需要生成式插画：顶部仅有约 230px 的版本标题，直接进入实际有内容的功能或修复区。这样版式与中文文字最稳定。用户明确想增加插画时，可用 imagegen 只生成无字装饰素材，再用脚本绘制全部文字与真实头像；不能让装饰占去首屏。
+
+`scan` 只用 Python 标准库；`render` 需要 Pillow 和可用的中文系统字体。缺依赖时先查看项目已有运行环境，不为一次制图修改生产依赖。GitHub 头像获取是公开只读请求；失败时给出具体账号和原因。

.agents/skills/release-announcement/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "版本发布长图"
+  short_description: "从最新 release note 生成微信群短文案、长图和贡献者头像"
+  default_prompt: "用 $release-announcement 读取本项目最新发布说明，制作微信群短文案和发布长图。"

.agents/skills/release-announcement/references/editorial-brief.md

@@ -0,0 +1,9 @@
+# 发布图与群文案提示词
+
+使用最新 release markdown 的**中文原文**作为唯一内容依据。写给已经在使用 Claude Code Haha 的微信群用户，突出他们最能感受到的变化，避免把每条 commit 都挤进图里。
+
+**长图视觉提示词：**1080px 宽、适合手机纵向阅读的简体中文版本更新图。顶部深炭色 Hero 约 230px，只放产品名、版本号和「新版本发布」，无插画、无大图、无大块留白；紧接着进入本版实际有内容的功能或修复区。主体为暖白纸色，深色标题、陶土红小标记、简洁双列功能卡、单列修复清单。纯修复版不要出现「这次新增」或空功能卡。贡献者区展示从 GitHub 实际取得的圆形头像、准确 ID 和对应贡献。正文由确定性字体排版完成，不能依赖生成模型拼写文字或绘制头像。按内容自然增长高度，不压缩字号来塞入更多条目。
+
+**提炼提示词：**从发布说明选最多 4 项核心功能、最多 9 项用户能理解的修复；没有的类别保持为空。每项标题尽量不超过 14 个汉字，说明用一句话讲清“现在能做什么”或“之前的问题如何改善”。优先考虑群友直接感受到的变化，避免用开发者内部修复挤掉有用的用户功能。保留必要条件，例如“在设置中开启后”、超时选项、适用平台、临时性或签名状态。模型更新可合并为一个区块。未署名的修复不强行归给贡献者。
+
+**群文案提示词：**用自然、简短的群聊语气，最多三句话；写出版本号和最重要的两项变化，最后引导看图。有本版署名贡献者时感谢代码贡献；没有时感谢反馈与支持。只有用户明确说这些贡献者是群成员时才称他们为“群友”。不要把图内清单再抄一遍；不要虚构下载链接或给未核实的 GitHub ID 加 `@`。

.agents/skills/release-announcement/references/poster-spec.md

@@ -0,0 +1,29 @@
+# `poster-spec.json` 格式
+
+运行 `scan` 后，以最新 release note 填写 JSON。字段：
+
+```json
+{
+  "version": "0.6.7",
+  "product": "Claude Code Haha",
+  "features": [
+    {"title": "侧边对话", "body": "输入 /btw，在右侧开启独立的临时对话，不打断主任务。"}
+  ],
+  "extras": [
+    {"title": "模型与供应商", "headline": "新版模型与映射", "body": "依据本版说明压缩成一句话。"}
+  ],
+  "fixes": [
+    {"title": "超大会话打不开", "body": "超过 20MB 的历史可以继续打开。"}
+  ],
+  "contributors": [
+    {"login": "example-user", "contribution": "此人本次被署名的贡献"}
+  ],
+  "install_note": "仅在发布说明明确写有安装提醒时填写。",
+  "footer_url": "github.com/NanmiCoder/cc-haha"
+}
+```
+
+- `version`、`features`、`fixes`、`contributors` 必填；`extras`、`install_note`、`footer_url` 可省略。纯修复版填 `"features": []`，纯功能版填 `"fixes": []`，两者不能同时为空。没有署名贡献者时 `contributors` 为 `[]`。
+- `render --scan` 会校验版本及贡献者 ID 必须是 release note 中文部分的明确署名候选。普通提及不算署名；不要把示例 ID 写入真实结果。
+- 测试或离线排版时，贡献者可以另给 `avatar_path` 并显式传 `--allow-offline-avatars`；正式生成禁止该选项，必须实时核验公开 GitHub 账号与头像。
+- 正式图保留所有已核验、已署名的贡献者；如果名字很多，可以缩短每人的贡献概述，不能遗漏账号或用假头像填位。

.agents/skills/release-announcement/scripts/.gitignore

@@ -0,0 +1,2 @@
+__pycache__/
+*.pyc

.agents/skills/release-announcement/scripts/release_poster.py

@@ -0,0 +1,392 @@
+#!/usr/bin/env python3
+"""Find a release note and render a compact, text-accurate WeChat poster."""
+
+import argparse
+import hashlib
+import json
+import re
+import subprocess
+import sys
+from pathlib import Path
+from urllib.error import HTTPError, URLError
+from urllib.parse import quote
+from urllib.request import Request, urlopen
+
+
+VERSION_RE = re.compile(r"^release-notes/v(\d+)\.(\d+)\.(\d+)\.md$")
+LOGIN_RE = re.compile(r"(?<![A-Za-z0-9._%+-])@([A-Za-z0-9](?:[A-Za-z0-9-]{0,37}[A-Za-z0-9])?)")
+RELEASE_HEADING_RE = re.compile(r"(?m)^#\s+Claude Code Haha v\d+\.\d+\.\d+\s*$")
+CLOSING_PUNCTUATION = set("，。、；：！？）》」』】〕”’％")
+OPENING_PUNCTUATION = set("（《「『【〔“‘")
+
+
+def git(repo, *args):
+    return subprocess.run(
+        ["git", "-C", str(repo), *args],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+
+
+def release_candidates(repo):
+    notes = repo / "release-notes"
+    candidates = []
+    if notes.is_dir():
+        for path in notes.glob("v*.md"):
+            relative = path.relative_to(repo).as_posix()
+            match = VERSION_RE.fullmatch(relative)
+            if match:
+                candidates.append((tuple(map(int, match.groups())), 4, "working-tree", relative))
+
+    for rank, ref in ((3, "HEAD"), (2, "main"), (1, "origin/main")):
+        result = git(repo, "ls-tree", "-r", "--name-only", ref, "--", "release-notes")
+        if result.returncode:
+            continue
+        for relative in result.stdout.splitlines():
+            match = VERSION_RE.fullmatch(relative)
+            if match:
+                candidates.append((tuple(map(int, match.groups())), rank, ref, relative))
+    return candidates
+
+
+def chinese_release_section(markdown):
+    headings = list(RELEASE_HEADING_RE.finditer(markdown))
+    if len(headings) >= 2:
+        sections = [
+            markdown[heading.start():(headings[index + 1].start() if index + 1 < len(headings) else len(markdown))]
+            for index, heading in enumerate(headings)
+        ]
+    else:
+        sections = [markdown]
+    # Older releases put English first and Chinese in <details>; others invert
+    # that order or repeat the H1 without <details>.
+    candidates = []
+    for section in sections:
+        candidates.extend(re.split(r"(?i)<details(?:\s[^>]*)?>", section))
+    selected = max(candidates, key=lambda part: len(re.findall(r"[\u4e00-\u9fff]", part)))
+    if len(re.findall(r"[\u4e00-\u9fff]", selected)) < 10:
+        raise ValueError("发布说明里未找到足够的中文正文，请人工核对语言版本")
+    return re.sub(r"(?is)</?details[^>]*>|<summary[^>]*>.*?</summary>", "", selected).strip() + "\n"
+
+
+def is_explicit_credit(line, login):
+    escaped = re.escape(login)
+    if re.search(rf"\*\*@{escaped}\*\*", line, re.IGNORECASE):
+        return True
+    return bool(re.search(rf"(?:感谢|致谢|代码贡献者|贡献者)[^\n@]{{0,70}}@{escaped}\b", line, re.IGNORECASE))
+
+
+def wrap_lines(value, width, measure):
+    result, current = [], ""
+    # Latin model names and IDs stay together; CJK wraps character by character.
+    tokens = re.findall(r"[A-Za-z0-9][A-Za-z0-9._/+:-]*|.", str(value), flags=re.DOTALL)
+    for token in tokens:
+        if token == "\n":
+            result.append(current)
+            current = ""
+            continue
+        trial = current + token
+        if current and measure(trial) > width:
+            if token in CLOSING_PUNCTUATION:
+                tail_match = re.search(r"(?:[A-Za-z0-9._/+:-]+|.)$", current)
+                tail = tail_match.group(0) if tail_match else ""
+                prefix = current[:-len(tail)].rstrip() if tail else ""
+                if prefix:
+                    result.append(prefix)
+                    current = tail + token
+                else:
+                    current = trial
+            elif current[-1] in OPENING_PUNCTUATION and len(current) > 1:
+                result.append(current[:-1].rstrip())
+                current = current[-1] + token
+            else:
+                result.append(current.rstrip())
+                current = token.lstrip()
+        else:
+            current = trial
+    result.append(current)
+    return result
+
+
+def scan(repo, requested_version=None):
+    candidates = release_candidates(repo)
+    if requested_version:
+        requested = requested_version.removeprefix("v")
+        if not re.fullmatch(r"\d+\.\d+\.\d+", requested):
+            raise ValueError(f"无效版本号：{requested_version}")
+        target = tuple(map(int, requested.split(".")))
+        candidates = [candidate for candidate in candidates if candidate[0] == target]
+    if not candidates:
+        label = requested_version or "最新版本"
+        raise ValueError(f"未找到 {label} 的 release-notes Markdown")
+    version, _, source, relative = max(candidates)
+    if source == "working-tree":
+        markdown = (repo / relative).read_text(encoding="utf-8")
+    else:
+        result = git(repo, "show", f"{source}:{relative}")
+        if result.returncode:
+            raise ValueError(f"无法从 {source} 读取 {relative}: {result.stderr.strip()}")
+        markdown = result.stdout
+
+    chinese = chinese_release_section(markdown)
+    mentions = {}
+    for line in chinese.splitlines():
+        for match in LOGIN_RE.finditer(line):
+            login = match.group(1)
+            key = login.lower()
+            entry = mentions.setdefault(key, {"login": login, "lines": [], "credit_candidate": False})
+            if line not in entry["lines"]:
+                entry["lines"].append(line.strip())
+            if is_explicit_credit(line, login):
+                entry["credit_candidate"] = True
+
+    return {
+        "version": ".".join(map(str, version)),
+        "source": source,
+        "path": relative,
+        "sha256": hashlib.sha256(chinese.encode("utf-8")).hexdigest(),
+        "chinese_markdown": chinese,
+        "mentions": list(mentions.values()),
+    }
+
+
+def font_paths(bold):
+    mac = "/System/Library/Fonts/STHeiti Medium.ttc" if bold else "/System/Library/Fonts/STHeiti Light.ttc"
+    linux = "/usr/share/fonts/opentype/noto/NotoSansCJK-Bold.ttc" if bold else "/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc"
+    windows = "C:/Windows/Fonts/msyhbd.ttc" if bold else "C:/Windows/Fonts/msyh.ttc"
+    return [mac, linux, windows, "/System/Library/Fonts/Hiragino Sans GB.ttc"]
+
+
+def choose_font_path(bold):
+    for candidate in font_paths(bold):
+        if Path(candidate).is_file():
+            return candidate
+    raise RuntimeError("缺少可用的中文系统字体；请提供宋体、黑体或 Noto CJK 字体")
+
+
+def check_spec(spec, scan_data, allow_offline_avatars=False):
+    if not isinstance(spec, dict):
+        raise ValueError("poster spec 必须是 JSON 对象")
+    if spec.get("version") != scan_data.get("version"):
+        raise ValueError("poster spec 的 version 与最新 release note 不一致")
+    for field in ("features", "fixes", "contributors"):
+        if not isinstance(spec.get(field), list):
+            raise ValueError(f"poster spec 缺少数组字段 {field}")
+    if not spec["features"] and not spec["fixes"]:
+        raise ValueError("至少需要一个主要功能或修复项")
+    if len(spec["features"]) > 6 or len(spec["fixes"]) > 12:
+        raise ValueError("内容过多：最多 6 项功能、12 项修复，请先提炼")
+    mention_ids = {m["login"].lower() for m in scan_data.get("mentions", []) if m.get("credit_candidate") is True}
+    seen = set()
+    for person in spec["contributors"]:
+        login = person.get("login", "")
+        if not isinstance(login, str) or not re.fullmatch(r"[A-Za-z0-9](?:[A-Za-z0-9-]{0,37}[A-Za-z0-9])?", login):
+            raise ValueError(f"无效 GitHub ID: {login!r}")
+        if login.lower() not in mention_ids:
+            raise ValueError(f"贡献者 @{login} 未在 release note 中文部分明确署名")
+        if login.lower() in seen:
+            raise ValueError(f"重复的贡献者 @{login}")
+        seen.add(login.lower())
+        if not person.get("contribution"):
+            raise ValueError(f"缺少 @{login} 的贡献概述")
+        if person.get("avatar_path") and not allow_offline_avatars:
+            raise ValueError("avatar_path 仅供显式离线测试使用，正式成图必须核验 GitHub 头像")
+    for field in ("features", "fixes"):
+        for item in spec[field]:
+            if not item.get("title") or not item.get("body"):
+                raise ValueError(f"{field} 每项都需要 title 和 body")
+
+
+def get_avatar(person, spec_path, avatar_dir, allow_offline_avatars=False):
+    from PIL import Image
+
+    login = person["login"]
+    if person.get("avatar_path"):
+        if not allow_offline_avatars:
+            raise ValueError("正式成图不能使用 avatar_path")
+        path = Path(person["avatar_path"])
+        if not path.is_absolute():
+            path = spec_path.parent / path
+        return Image.open(path).convert("RGB")
+
+    request = Request(
+        "https://api.github.com/users/" + quote(login),
+        headers={"Accept": "application/vnd.github+json", "User-Agent": "cc-haha-release-announcement"},
+    )
+    try:
+        with urlopen(request, timeout=20) as response:
+            profile = json.load(response)
+        if profile.get("login", "").lower() != login.lower():
+            raise ValueError(f"GitHub 返回的账号与 @{login} 不符")
+        url = profile.get("avatar_url")
+        if not url:
+            raise ValueError(f"@{login} 没有公开头像地址")
+        image_request = Request(url + "&s=256", headers={"User-Agent": "cc-haha-release-announcement"})
+        with urlopen(image_request, timeout=20) as response:
+            content = response.read()
+    except (HTTPError, URLError, TimeoutError) as exc:
+        raise RuntimeError(f"获取 @{login} 的 GitHub 头像失败：{exc}") from exc
+    avatar_dir.mkdir(parents=True, exist_ok=True)
+    path = avatar_dir / (login + ".png")
+    path.write_bytes(content)
+    return Image.open(path).convert("RGB")
+
+
+def render(spec, scan_data, spec_path, output, allow_offline_avatars=False):
+    from PIL import Image, ImageDraw, ImageFont, ImageOps
+
+    check_spec(spec, scan_data, allow_offline_avatars)
+    regular_path, bold_path = choose_font_path(False), choose_font_path(True)
+    def face(size, bold=False):
+        return ImageFont.truetype(bold_path if bold else regular_path, size)
+
+    W, MAX_H = 1080, 16000
+    paper, ink, muted, rust, line = "#F8F5EE", "#25211D", "#746B62", "#A84D35", "#DED5C9"
+    canvas = Image.new("RGB", (W, MAX_H), paper)
+    draw = ImageDraw.Draw(canvas)
+
+    def lines(value, text_face, width):
+        return wrap_lines(value, width, lambda text: draw.textlength(text, font=text_face))
+
+    def write(x, y, value, text_face, color, width, leading=1.38):
+        step = round(text_face.size * leading)
+        for text_line in lines(value, text_face, width):
+            draw.text((x, y), text_line, font=text_face, fill=color)
+            y += step
+        return y
+
+    # Compact hero: the first release content appears in the first phone viewport.
+    draw.rectangle((0, 0, W, 230), fill=ink)
+    product = str(spec.get("product") or "Claude Code Haha").upper()
+    write(64, 40, f"{product}  ·  v{spec['version']}", face(29, True), "#D6B99E", 950)
+    draw.text((64, 99), "新版本发布", font=face(63, True), fill="#F4EFE4")
+    draw.line((64, 207, 1016, 207), fill=rust, width=4)
+
+    y = 286
+    features = spec["features"]
+    if features:
+        draw.text((64, y), "这次新增", font=face(58, True), fill=ink)
+        draw.text((68, y + 80), "本次最值得关注的变化", font=face(30), fill=muted)
+        y += 150
+        cols = 1 if len(features) == 1 else 2
+        gap_x, gap_y = 28, 28
+        for start in range(0, len(features), cols):
+            row = features[start:start + cols]
+            card_w = 952 if len(row) == 1 else 461
+            dimensions = []
+            for item in row:
+                title_lines = lines(item["title"], face(37, True), card_w - 60)
+                body_lines = lines(item["body"], face(30), card_w - 60)
+                body_top = 87 + len(title_lines) * 45 + 24
+                dimensions.append((title_lines, body_lines, body_top))
+            minimum_height = 240 if len(row) == 1 else 340
+            height = max(minimum_height, *(body_top + len(body_lines) * 42 + 35 for _, body_lines, body_top in dimensions))
+            for col, (item, (_, _, body_top)) in enumerate(zip(row, dimensions)):
+                x = 64 + col * (card_w + gap_x)
+                draw.rounded_rectangle((x, y, x + card_w, y + height), radius=26, fill="#FFFFFF", outline=line, width=2)
+                draw.text((x + 30, y + 28), f"{start + col + 1:02d}", font=face(28, True), fill=rust)
+                write(x + 30, y + 87, item["title"], face(37, True), ink, card_w - 60, 1.22)
+                write(x + 30, y + body_top, item["body"], face(30), muted, card_w - 60, 1.4)
+            y += height + gap_y
+        y += 66
+
+    for extra in spec.get("extras", []):
+        if not all(extra.get(key) for key in ("title", "headline", "body")):
+            raise ValueError("extras 每项需要 title、headline、body")
+        draw.text((64, y), extra["title"], font=face(54, True), fill=ink)
+        y += 90
+        head_lines = lines(extra["headline"], face(34, True), 890)
+        body_lines = lines(extra["body"], face(30), 880)
+        box_h = 30 + len(head_lines) * 45 + 20 + len(body_lines) * 40 + 30
+        draw.rounded_rectangle((64, y, 1016, y + box_h), radius=25, fill="#EEE4D7")
+        write(94, y + 30, extra["headline"], face(34, True), ink, 890, 1.32)
+        write(94, y + 30 + len(head_lines) * 45 + 20, extra["body"], face(30), muted, 880, 1.38)
+        y += box_h + 80
+
+    if spec["fixes"]:
+        draw.text((64, y), "重点修复", font=face(58, True), fill=ink)
+        draw.text((68, y + 80), "那些真正影响使用的问题", font=face(30), fill=muted)
+        y += 160
+        for index, item in enumerate(spec["fixes"]):
+            row_y = y
+            draw.ellipse((67, y + 16, 83, y + 32), fill=rust)
+            title_end = write(107, y, item["title"], face(34, True), ink, 870, 1.2)
+            body_end = write(107, title_end + 12, item["body"], face(30), muted, 870, 1.36)
+            y = max(row_y + 116, body_end + 24)
+            if index < len(spec["fixes"]) - 1:
+                draw.line((107, y - 13, 1016, y - 13), fill=line, width=2)
+        y += 56
+
+    people = spec["contributors"]
+    if people:
+        draw.text((64, y), "感谢社区贡献者", font=face(54, True), fill=ink)
+        draw.text((67, y + 78), "头像与 GitHub ID 来自公开主页", font=face(28), fill=muted)
+        y += 150
+        avatar_dir = output.parent / "avatars"
+        avatars = [(person, get_avatar(person, spec_path, avatar_dir, allow_offline_avatars)) for person in people]
+        for person, photo in avatars:
+            contribution_lines = lines(person["contribution"], face(24), 805)
+            row_h = max(116, 75 + len(contribution_lines) * 31)
+            draw.rounded_rectangle((64, y, 1016, y + row_h), radius=20, fill="#FFFFFF", outline=line, width=2)
+            square = ImageOps.fit(photo, (78, 78), method=Image.Resampling.LANCZOS)
+            mask = Image.new("L", (78, 78))
+            ImageDraw.Draw(mask).ellipse((0, 0, 77, 77), fill=255)
+            canvas.paste(square, (86, y + 19), mask)
+            draw.text((184, y + 23), "@" + person["login"], font=face(30, True), fill=ink)
+            write(184, y + 66, person["contribution"], face(24), muted, 805)
+            y += row_h + 16
+        y += 20
+
+    if spec.get("install_note"):
+        note_lines = lines(spec["install_note"], face(27), 885)
+        box_h = 77 + len(note_lines) * 38 + 30
+        draw.rounded_rectangle((64, y, 1016, y + box_h), radius=25, fill=ink)
+        draw.text((96, y + 28), "安装提示", font=face(31, True), fill="#F5D3B4")
+        write(96, y + 76, spec["install_note"], face(27), "#F4EFE4", 885)
+        y += box_h + 68
+
+    draw.text((64, y), f"{spec.get('product') or 'Claude Code Haha'}  ·  v{spec['version']}", font=face(28, True), fill=ink)
+    if spec.get("footer_url"):
+        draw.text((64, y + 48), spec["footer_url"], font=face(27), fill=muted)
+    y += 137
+    if y > MAX_H:
+        raise ValueError(f"海报高度 {y}px 超出上限，请减少内容")
+    output.parent.mkdir(parents=True, exist_ok=True)
+    canvas.crop((0, 0, W, y)).save(output, optimize=True)
+    return (W, y)
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    commands = parser.add_subparsers(dest="command", required=True)
+    scan_cmd = commands.add_parser("scan", help="查找最新 release note，并输出中文内容与署名候选")
+    scan_cmd.add_argument("--repo", type=Path, default=Path.cwd())
+    scan_cmd.add_argument("--version", help="指定旧版本进行复测，例如 0.6.6；默认选最新版本")
+    scan_cmd.add_argument("--out", type=Path, required=True)
+    render_cmd = commands.add_parser("render", help="从经过审核的 JSON 绘制微信群长图")
+    render_cmd.add_argument("--scan", type=Path, required=True)
+    render_cmd.add_argument("--spec", type=Path, required=True)
+    render_cmd.add_argument("--out", type=Path, required=True)
+    render_cmd.add_argument("--allow-offline-avatars", action="store_true", help="仅测试用：允许 avatar_path 跳过 GitHub 请求")
+    args = parser.parse_args()
+    try:
+        if args.command == "scan":
+            data = scan(args.repo.resolve(), args.version)
+            args.out.parent.mkdir(parents=True, exist_ok=True)
+            args.out.write_text(json.dumps(data, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
+            credited = sum(mention["credit_candidate"] for mention in data["mentions"])
+            print(f"v{data['version']} | {data['source']}:{data['path']} | {credited} credits / {len(data['mentions'])} mentions | {args.out}")
+        else:
+            scan_data = json.loads(args.scan.read_text(encoding="utf-8"))
+            spec = json.loads(args.spec.read_text(encoding="utf-8"))
+            width, height = render(spec, scan_data, args.spec, args.out, args.allow_offline_avatars)
+            print(f"{args.out} | {width}x{height}")
+    except (ValueError, RuntimeError, OSError, KeyError, json.JSONDecodeError) as exc:
+        print(f"error: {exc}", file=sys.stderr)
+        return 1
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

.agents/skills/release-announcement/scripts/test_release_poster.py

@@ -0,0 +1,83 @@
+import tempfile
+import unittest
+from pathlib import Path
+
+from release_poster import check_spec, chinese_release_section, scan, wrap_lines
+
+
+class ReleasePosterTest(unittest.TestCase):
+    def test_latest_semver_and_chinese_credits(self):
+        with tempfile.TemporaryDirectory() as directory:
+            repo = Path(directory)
+            notes = repo / "release-notes"
+            notes.mkdir()
+            (notes / "v0.6.9.md").write_text("# v0.6.9\n本次版本修复桌面端加载问题 **@latest-user**\n<details>\n@english-only\n", encoding="utf-8")
+            (notes / "v0.6.8.md").write_text("# v0.6.8\n本次版本修复桌面端加载问题 **@old-user**\n", encoding="utf-8")
+            result = scan(repo)
+            self.assertEqual(result["version"], "0.6.9")
+            self.assertEqual([person["login"] for person in result["mentions"]], ["latest-user"])
+            older = scan(repo, "v0.6.8")
+            self.assertEqual(older["version"], "0.6.8")
+            self.assertEqual([person["login"] for person in older["mentions"]], ["old-user"])
+
+    def test_chinese_section_in_either_order_and_without_details(self):
+        title = "# Claude Code Haha v0.5.4\n"
+        english_first = title + "English changes here.\n<details>\n<summary>中文版本</summary>\n" + title + "这里是中文修复内容和版本更新。\n</details>"
+        chinese_first = title + "这里是中文修复内容和版本更新。\n<details>\n<summary>English</summary>\n" + title + "English changes here.\n</details>"
+        no_details = title + "这里是中文修复内容和版本更新。\n" + title + "English changes here.\n"
+        for markdown in (english_first, chinese_first, no_details):
+            selected = chinese_release_section(markdown)
+            self.assertIn("这里是中文修复内容", selected)
+            self.assertNotIn("English changes", selected)
+
+    def test_incidental_mention_is_not_contributor_credit(self):
+        with tempfile.TemporaryDirectory() as directory:
+            repo = Path(directory)
+            notes = repo / "release-notes"
+            notes.mkdir()
+            (notes / "v0.2.0.md").write_text(
+                "# Claude Code Haha v0.2.0\n本次修复桌面问题。\n- PR 模板默认要求 @dosubot review。\n- 修复连接问题 **@actual-author**（#42）。\n",
+                encoding="utf-8",
+            )
+            result = scan(repo)
+            candidates = {person["login"]: person["credit_candidate"] for person in result["mentions"]}
+            self.assertEqual(candidates, {"dosubot": False, "actual-author": True})
+            spec = {
+                "version": "0.2.0", "features": [], "fixes": [{"title": "修复", "body": "说明"}],
+                "contributors": [{"login": "dosubot", "contribution": "代码贡献"}],
+            }
+            with self.assertRaisesRegex(ValueError, "未在 release note 中文部分明确署名"):
+                check_spec(spec, result)
+
+    def test_cannot_credit_unmentioned_account(self):
+        source = {"version": "0.6.7", "mentions": [{"login": "actual-author", "credit_candidate": True}]}
+        spec = {
+            "version": "0.6.7",
+            "features": [{"title": "功能", "body": "说明"}],
+            "fixes": [{"title": "修复", "body": "说明"}],
+            "contributors": [{"login": "different-author", "contribution": "贡献"}],
+        }
+        with self.assertRaisesRegex(ValueError, "未在 release note 中文部分明确署名"):
+            check_spec(spec, source)
+
+    def test_offline_avatar_requires_explicit_test_mode(self):
+        source = {"version": "0.6.7", "mentions": [{"login": "actual-author", "credit_candidate": True}]}
+        spec = {
+            "version": "0.6.7", "features": [], "fixes": [{"title": "修复", "body": "说明"}],
+            "contributors": [{"login": "actual-author", "contribution": "修复", "avatar_path": "fake.png"}],
+        }
+        with self.assertRaisesRegex(ValueError, "仅供显式离线测试"):
+            check_spec(spec, source)
+        check_spec(spec, source, allow_offline_avatars=True)
+
+    def test_chinese_punctuation_does_not_start_a_wrapped_line(self):
+        width = 4
+        measure = len
+        for value in ("这是测试；继续", "这是测试。继续", "这是测试、继续", "测试「功能描述"):
+            lines = wrap_lines(value, width, measure)
+            self.assertTrue(all(not line.startswith(("；", "。", "、")) for line in lines[1:]))
+            self.assertTrue(all(not line.endswith("「") for line in lines[:-1]))
+
+
+if __name__ == "__main__":
+    unittest.main()