增加 Release 自动同步 workflow

对应源仓 think-twice 的 d80e546。副本内容与源仓一致。

SKILL.md:blocker 五项改为显式有序;读者表格不再按"top-left cell"这类位置引用;
style.md 的加载条件(含 code-heavy 跳过那个例外)从 style.md 搬进 SKILL.md 正文。

style.md:删掉开头写给维护者的说明,只留用途和用法。

.dist-sync/sources.json

@@ -0,0 +1,9 @@
+{
+  "sources": [
+    {
+      "name": "think-twice",
+      "repository": "agent-skills/think-twice",
+      "asset": "think-twice.zip"
+    }
+  ]
+}

.dist-sync/versions.json

@@ -0,0 +1 @@
+{}

.gitea/workflows/sync-releases.yml

@@ -0,0 +1,92 @@
+name: 同步 skill Releases
+
+on:
+  schedule:
+    - cron: "17 * * * *"
+  workflow_dispatch:
+
+concurrency:
+  group: sync-skill-releases
+  cancel-in-progress: false
+
+permissions:
+  contents: write
+
+jobs:
+  sync:
+    runs-on: ubuntu-latest
+    steps:
+      - name: 检出分发仓
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+
+      - name: 检查同步凭据
+        env:
+          SOURCE_REPOS_TOKEN: ${{ secrets.AGENT_SKILL_BOT }}
+        run: |
+          set -eu
+          test -n "$SOURCE_REPOS_TOKEN" || {
+            echo "缺少组织 Actions secret：AGENT_SKILL_BOT"
+            exit 1
+          }
+
+      - name: 同步 latest Releases
+        env:
+          GITEA_SERVER_URL: ${{ github.server_url }}
+          SOURCE_REPOS_TOKEN: ${{ secrets.AGENT_SKILL_BOT }}
+        run: python3 scripts/sync-releases.py
+
+      - name: 校验分发仓
+        run: |
+          set -eu
+          python3 scripts/validate-dist.py
+          NO_COLOR=1 npx --yes skills@1.5.22 add . --list | tee /tmp/skills-list.txt
+          python3 - <<'PY'
+          from pathlib import Path
+
+          expected = {
+              "baimiao-table-ocr",
+              "disk-master",
+              "everything-file-search",
+              "teach-me",
+              "think-twice",
+          }
+          output = Path("/tmp/skills-list.txt").read_text(encoding="utf-8")
+          lines = {line.strip(" │└├─") for line in output.splitlines()}
+          missing = sorted(expected - lines)
+          if missing:
+              raise SystemExit(f"skills CLI 未发现：{', '.join(missing)}")
+          print(f"skills CLI 已发现全部 {len(expected)} 个 skill")
+          PY
+
+      - name: 提交同步结果
+        env:
+          GITEA_TOKEN: ${{ secrets.GITEA_TOKEN }}
+          GITEA_SERVER_URL: ${{ github.server_url }}
+        run: |
+          set -eu
+          if git diff --quiet -- .dist-sync skills; then
+            echo "latest Releases 没有产生内容变化"
+            exit 0
+          fi
+
+          git config user.name "Agent Skills Bot"
+          git config user.email "agent-skills-bot@im0o.top"
+          git add .dist-sync skills
+          git commit -m "同步 latest skill releases"
+
+          auth=$(printf 'x-access-token:%s' "$GITEA_TOKEN" | base64 | tr -d '\n')
+          server_url=${GITEA_SERVER_URL%/}
+          git_auth() {
+            git -c "http.${server_url}/.extraheader=Authorization: Basic $auth" "$@"
+          }
+
+          git_auth fetch origin main
+          test "$(git rev-parse HEAD^)" = "$(git rev-parse origin/main)" || {
+            echo "origin/main 已前进，本次不自动变基或强制推送"
+            exit 1
+          }
+
+          git_auth push origin HEAD:main

README.md

@@ -1,13 +1,14 @@
 # agent-skills
 
-一次装齐 im0o 的 agent skill。
+从一个入口按用途选择并安装 im0o 的 agent skill。
 
 ```bash
-npx skills add ssh://gitea/agent-skills/skills
+npx skills add ssh://gitea/agent-skills/skills.git
 ```
 
-会出现一个多选列表,按用途分成三组,勾选需要的即可。仓库是私有的,`skills add`
-用系统 git 克隆,所以只要你的 SSH key 能访问就能装,不需要额外配置凭据。
+会出现一个按用途分组的多选列表。可以选择整组，也可以只勾选其中几个 skill。仓库是
+私有的，`skills add` 使用系统 Git 克隆；本机的 SSH key 能访问该仓库即可，不需要另外
+配置安装凭据。
 
 ## 目录里有什么
 
@@ -37,10 +38,13 @@ npx skills add ssh://gitea/agent-skills/skills
 
 | skill | 源码仓 | 仓内路径 |
 |---|---|---|
-| `think-twice` | `agent-skills/think-twice` | 仓根 |
+| `think-twice` | `agent-skills/think-twice` | `skill/` |
 | `teach-me` | `agent-skills/teach-me` | 仓根 |
 | `disk-master` | `agent-skills/disk-master-skill` | `skill/` |
 | `everything-file-search` | `agent-skills/everything-skill` | `everything-file-search/` |
 | `baimiao-table-ocr` | `agent-skills/img2excel-skill` | `skills/baimiao-table-ocr/` |
 
-改动在源码仓做,这个仓只承载分发副本。同步目前是手工的,自动化还没做。
+改动在源码仓完成，分发仓只承载可安装副本。源码仓发布正式 `v*` tag 并生成 Release
+ZIP 后，分发仓按 `.dist-sync/sources.json` 白名单同步 latest Release。当前只自动同步
+`think-twice`；其余四个 skill 具备标准 Release 后再逐项加入。白名单来源尚无 latest
+Release 时会跳过并保留现有副本。

scripts/sync-releases.py

@@ -0,0 +1,440 @@
+#!/usr/bin/env python3
+"""Synchronize allowlisted skills from their latest Gitea Release assets."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import re
+import shutil
+import stat
+import sys
+import tempfile
+import urllib.error
+import urllib.parse
+import urllib.request
+import zipfile
+from pathlib import Path, PurePosixPath
+from typing import Any
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+DEFAULT_SOURCES = REPO_ROOT / ".dist-sync" / "sources.json"
+DEFAULT_VERSIONS = REPO_ROOT / ".dist-sync" / "versions.json"
+MAX_ASSET_BYTES = 250 * 1024 * 1024
+MAX_API_BYTES = 5 * 1024 * 1024
+MAX_ARCHIVE_FILES = 10_000
+MAX_ARCHIVE_DEPTH = 32
+MAX_ARCHIVE_PATH_LENGTH = 1024
+TAG_PATTERN = re.compile(r"v[0-9]+(?:\.[0-9]+)*")
+
+
+class SyncError(RuntimeError):
+    """A source exists but cannot be synchronized safely."""
+
+
+def log(message: str) -> None:
+    print(message, flush=True)
+
+
+def read_json(path: Path) -> Any:
+    try:
+        return json.loads(path.read_text(encoding="utf-8"))
+    except (OSError, json.JSONDecodeError) as error:
+        raise SyncError(f"无法读取 JSON：{path}: {error}") from error
+
+
+def write_json_atomic(path: Path, value: Any) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    encoded = (json.dumps(value, ensure_ascii=False, indent=2, sort_keys=True) + "\n").encode("utf-8")
+    temporary: Path | None = None
+    try:
+        with tempfile.NamedTemporaryFile(prefix=f".{path.name}.", dir=path.parent, delete=False) as output:
+            temporary = Path(output.name)
+            output.write(encoded)
+            output.flush()
+            os.fsync(output.fileno())
+        os.replace(temporary, path)
+    except OSError as error:
+        if temporary is not None:
+            temporary.unlink(missing_ok=True)
+        raise SyncError(f"无法原子更新 JSON：{path}: {error}") from error
+
+
+def validate_source(raw: Any) -> dict[str, str]:
+    if not isinstance(raw, dict):
+        raise SyncError("sources.json 的每个来源必须是对象")
+
+    source: dict[str, str] = {}
+    for field in ("name", "repository", "asset"):
+        value = raw.get(field)
+        if not isinstance(value, str) or not value.strip():
+            raise SyncError(f"来源缺少有效字段：{field}")
+        source[field] = value.strip()
+
+    name = source["name"]
+    repository = source["repository"]
+    asset = source["asset"]
+    if not re.fullmatch(r"[a-z0-9]+(?:-[a-z0-9]+)*", name):
+        raise SyncError(f"无效 skill 名称：{name}")
+    owner, repo = repository.split("/", 1) if "/" in repository else ("", "")
+    repository_part = re.compile(r"[A-Za-z0-9_.-]+")
+    if not owner or not repo or not repository_part.fullmatch(owner) or not repository_part.fullmatch(repo):
+        raise SyncError(f"无效仓库名称：{repository}")
+    if Path(asset).name != asset:
+        raise SyncError(f"附件名称不能包含路径：{asset}")
+    return source
+
+
+def canonical_origin(url: str) -> tuple[str, str, int | None]:
+    parsed = urllib.parse.urlsplit(url)
+    scheme = parsed.scheme.lower()
+    hostname = parsed.hostname.lower() if parsed.hostname else ""
+    try:
+        port = parsed.port
+    except ValueError as error:
+        raise SyncError(f"URL 端口无效：{url}") from error
+    if scheme not in {"http", "https"} or not hostname:
+        raise SyncError(f"URL 必须是有效的 HTTP(S) 地址：{url}")
+    effective_port = port if port is not None else (443 if scheme == "https" else 80)
+    return scheme, hostname, effective_port
+
+
+class SameOriginRedirectHandler(urllib.request.HTTPRedirectHandler):
+    def __init__(self, allowed_origin: tuple[str, str, int | None]) -> None:
+        super().__init__()
+        self.allowed_origin = allowed_origin
+
+    def redirect_request(
+        self,
+        request: urllib.request.Request,
+        file_pointer: Any,
+        code: int,
+        message: str,
+        headers: Any,
+        new_url: str,
+    ) -> urllib.request.Request | None:
+        if canonical_origin(new_url) != self.allowed_origin:
+            raise SyncError(f"拒绝跨源重定向：{request.full_url} -> {new_url}")
+        return super().redirect_request(request, file_pointer, code, message, headers, new_url)
+
+
+def build_opener(server_url: str) -> tuple[urllib.request.OpenerDirector, tuple[str, str, int | None]]:
+    origin = canonical_origin(server_url)
+    return urllib.request.build_opener(SameOriginRedirectHandler(origin)), origin
+
+
+def request_json(url: str, token: str, opener: urllib.request.OpenerDirector) -> tuple[int, Any | None]:
+    request = urllib.request.Request(
+        url,
+        headers={
+            "Accept": "application/json",
+            "Authorization": f"token {token}",
+            "User-Agent": "agent-skills-dist-sync",
+        },
+    )
+    try:
+        with opener.open(request, timeout=30) as response:
+            body = response.read(MAX_API_BYTES + 1)
+            if len(body) > MAX_API_BYTES:
+                raise SyncError(f"Gitea API 响应超过 {MAX_API_BYTES} bytes：{url}")
+            try:
+                return response.status, json.loads(body)
+            except (UnicodeDecodeError, json.JSONDecodeError) as error:
+                raise SyncError(f"Gitea API 返回无效 JSON：{url}: {error}") from error
+    except urllib.error.HTTPError as error:
+        if error.code == 404:
+            return error.code, None
+        detail = error.read(2048).decode("utf-8", errors="replace").strip()
+        raise SyncError(f"Gitea API 返回 HTTP {error.code}: {detail or url}") from error
+    except SyncError:
+        raise
+    except OSError as error:
+        raise SyncError(f"请求 Gitea API 失败：{url}: {error}") from error
+
+
+def download(
+    url: str,
+    token: str,
+    destination: Path,
+    opener: urllib.request.OpenerDirector,
+    allowed_origin: tuple[str, str, int | None],
+) -> None:
+    if canonical_origin(url) != allowed_origin:
+        raise SyncError(f"附件下载地址与 Gitea 不同源，拒绝发送 token：{url}")
+    request = urllib.request.Request(
+        url,
+        headers={
+            "Accept": "application/octet-stream",
+            "Authorization": f"token {token}",
+            "User-Agent": "agent-skills-dist-sync",
+        },
+    )
+    try:
+        with opener.open(request, timeout=120) as response, destination.open("wb") as output:
+            declared_size = response.headers.get("Content-Length")
+            if declared_size:
+                try:
+                    if int(declared_size) > MAX_ASSET_BYTES:
+                        raise SyncError(f"附件超过 {MAX_ASSET_BYTES} bytes：{url}")
+                except ValueError as error:
+                    raise SyncError(f"附件返回无效 Content-Length：{declared_size!r}") from error
+
+            copied = 0
+            while chunk := response.read(1024 * 1024):
+                copied += len(chunk)
+                if copied > MAX_ASSET_BYTES:
+                    raise SyncError(f"附件超过 {MAX_ASSET_BYTES} bytes：{url}")
+                output.write(chunk)
+    except urllib.error.HTTPError as error:
+        detail = error.read(2048).decode("utf-8", errors="replace").strip()
+        raise SyncError(f"下载附件返回 HTTP {error.code}: {detail or url}") from error
+    except SyncError:
+        raise
+    except OSError as error:
+        raise SyncError(f"下载附件失败：{url}: {error}") from error
+
+
+def safe_extract(archive_path: Path, destination: Path) -> None:
+    try:
+        with zipfile.ZipFile(archive_path) as archive:
+            files = 0
+            total_size = 0
+            for info in archive.infolist():
+                normalized = info.filename.replace("\\", "/")
+                member = PurePosixPath(normalized)
+                if member.is_absolute() or ".." in member.parts:
+                    raise SyncError(f"ZIP 包含不安全路径：{info.filename}")
+                if not member.parts:
+                    continue
+                if len(normalized) > MAX_ARCHIVE_PATH_LENGTH or len(member.parts) > MAX_ARCHIVE_DEPTH:
+                    raise SyncError(f"ZIP 路径过长或目录过深：{info.filename}")
+
+                mode = info.external_attr >> 16
+                if stat.S_ISLNK(mode):
+                    raise SyncError(f"ZIP 不允许符号链接：{info.filename}")
+
+                target = destination.joinpath(*member.parts)
+                if not info.is_dir():
+                    files += 1
+                    if files > MAX_ARCHIVE_FILES:
+                        raise SyncError(f"ZIP 文件数量超过 {MAX_ARCHIVE_FILES}")
+                    total_size += info.file_size
+                    if total_size > MAX_ASSET_BYTES:
+                        raise SyncError("ZIP 解压后内容过大")
+                    target.parent.mkdir(parents=True, exist_ok=True)
+                    with archive.open(info) as source, target.open("wb") as output:
+                        shutil.copyfileobj(source, output)
+                    if mode & 0o111:
+                        target.chmod(target.stat().st_mode | 0o111)
+                else:
+                    target.mkdir(parents=True, exist_ok=True)
+
+            if files == 0:
+                raise SyncError("ZIP 中没有文件")
+    except zipfile.BadZipFile as error:
+        raise SyncError(f"附件不是有效 ZIP：{archive_path}") from error
+
+
+def frontmatter_name(skill_file: Path) -> str:
+    try:
+        lines = skill_file.read_text(encoding="utf-8-sig").splitlines()
+    except (OSError, UnicodeDecodeError) as error:
+        raise SyncError(f"无法读取 {skill_file}: {error}") from error
+
+    if not lines or lines[0].strip() != "---":
+        raise SyncError(f"SKILL.md 缺少 YAML frontmatter：{skill_file}")
+
+    for line in lines[1:]:
+        if line.strip() == "---":
+            break
+        if line.startswith("name:"):
+            return line.split(":", 1)[1].strip().strip("'\"")
+    raise SyncError(f"SKILL.md frontmatter 缺少 name：{skill_file}")
+
+
+def stage_source(
+    source: dict[str, str],
+    *,
+    server_url: str,
+    token: str,
+    current_versions: dict[str, str],
+    next_versions: dict[str, str],
+    temp_root: Path,
+    opener: urllib.request.OpenerDirector,
+    allowed_origin: tuple[str, str, int | None],
+) -> tuple[str, Path | None]:
+    name = source["name"]
+    repository = source["repository"]
+    encoded_repository = "/".join(urllib.parse.quote(part, safe="") for part in repository.split("/"))
+    latest_url = f"{server_url}/api/v1/repos/{encoded_repository}/releases/latest"
+
+    status, release = request_json(latest_url, token, opener)
+    if status == 404:
+        log(f"跳过 {name}：仓库不可见或尚无 latest Release")
+        return "skipped", None
+    if not isinstance(release, dict):
+        raise SyncError(f"{repository} 的 latest Release 响应无效")
+
+    tag = release.get("tag_name")
+    assets = release.get("assets")
+    if not isinstance(tag, str) or not TAG_PATTERN.fullmatch(tag):
+        raise SyncError(f"{repository} 的 latest Release tag 不符合 v* 正式版本约定：{tag!r}")
+    if not isinstance(assets, list):
+        raise SyncError(f"{repository} 的 latest Release 缺少附件列表")
+
+    matches = [asset for asset in assets if isinstance(asset, dict) and asset.get("name") == source["asset"]]
+    if len(matches) != 1:
+        raise SyncError(f"{repository}@{tag} 应恰好包含一个附件 {source['asset']}，实际为 {len(matches)} 个")
+
+    target = REPO_ROOT / "skills" / name
+    if current_versions.get(name) == tag and target.is_dir():
+        log(f"跳过 {name}：已经同步 {tag}")
+        return "unchanged", None
+
+    download_url = matches[0].get("browser_download_url")
+    if not isinstance(download_url, str) or not download_url:
+        raise SyncError(f"{repository}@{tag} 的附件缺少 browser_download_url")
+
+    work_dir = temp_root / name
+    extract_dir = work_dir / "content"
+    archive_path = work_dir / source["asset"]
+    extract_dir.mkdir(parents=True)
+    download(download_url, token, archive_path, opener, allowed_origin)
+    safe_extract(archive_path, extract_dir)
+
+    skill_file = extract_dir / "SKILL.md"
+    if not skill_file.is_file():
+        raise SyncError(f"{repository}@{tag} 的 ZIP 根目录缺少 SKILL.md")
+    actual_name = frontmatter_name(skill_file)
+    if actual_name != name:
+        raise SyncError(f"{repository}@{tag} 的 skill name 为 {actual_name!r}，预期 {name!r}")
+
+    next_versions[name] = tag
+    log(f"已准备 {name}：{current_versions.get(name, '未记录')} -> {tag}")
+    return "updated", extract_dir
+
+
+def apply_staged_updates(
+    staged: list[tuple[str, Path]],
+    current_versions: dict[str, str],
+    next_versions: dict[str, str],
+    versions_path: Path,
+) -> None:
+    if not staged:
+        return
+
+    backup_root = Path(tempfile.mkdtemp(prefix=".dist-sync-backup-", dir=REPO_ROOT))
+    applied: list[tuple[Path, Path | None]] = []
+    try:
+        for name, staged_dir in staged:
+            destination = REPO_ROOT / "skills" / name
+            replacement = destination.with_name(f".{name}.dist-sync-new")
+            if replacement.exists():
+                shutil.rmtree(replacement)
+            shutil.copytree(staged_dir, replacement)
+
+            backup: Path | None = None
+            if destination.exists():
+                backup = backup_root / name
+                destination.rename(backup)
+            try:
+                replacement.rename(destination)
+            except Exception:
+                if backup is not None and backup.exists() and not destination.exists():
+                    backup.rename(destination)
+                raise
+            applied.append((destination, backup))
+
+        write_json_atomic(versions_path, next_versions)
+    except Exception as error:
+        for destination, backup in reversed(applied):
+            if destination.exists():
+                shutil.rmtree(destination)
+            if backup is not None and backup.exists():
+                backup.rename(destination)
+        try:
+            write_json_atomic(versions_path, current_versions)
+        except SyncError as rollback_error:
+            raise SyncError(
+                f"应用同步结果失败；skill 目录已回滚，但 versions.json 回滚失败：{rollback_error}"
+            ) from error
+        if isinstance(error, SyncError):
+            raise
+        raise SyncError(f"应用同步结果失败，已回滚 skill 目录：{error}") from error
+    finally:
+        shutil.rmtree(backup_root, ignore_errors=True)
+        for name, _ in staged:
+            replacement = REPO_ROOT / "skills" / f".{name}.dist-sync-new"
+            if replacement.exists():
+                shutil.rmtree(replacement, ignore_errors=True)
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--sources", type=Path, default=DEFAULT_SOURCES)
+    parser.add_argument("--versions", type=Path, default=DEFAULT_VERSIONS)
+    parser.add_argument("--server-url", default=os.environ.get("GITEA_SERVER_URL", ""))
+    parser.add_argument("--token", default=os.environ.get("SOURCE_REPOS_TOKEN", ""))
+    args = parser.parse_args()
+
+    server_url = args.server_url.rstrip("/")
+    token = args.token.strip()
+    if not server_url:
+        raise SyncError("缺少 GITEA_SERVER_URL 或 --server-url")
+    if not token:
+        raise SyncError("缺少 SOURCE_REPOS_TOKEN 或 --token")
+    opener, allowed_origin = build_opener(server_url)
+
+    config = read_json(args.sources)
+    raw_sources = config.get("sources") if isinstance(config, dict) else None
+    if not isinstance(raw_sources, list) or not raw_sources:
+        raise SyncError("sources.json 必须包含非空 sources 数组")
+    sources = [validate_source(raw) for raw in raw_sources]
+    names = [source["name"] for source in sources]
+    if len(names) != len(set(names)):
+        raise SyncError("sources.json 包含重复的 skill name")
+
+    raw_versions = read_json(args.versions) if args.versions.exists() else {}
+    if not isinstance(raw_versions, dict) or not all(
+        isinstance(name, str) and isinstance(tag, str) and TAG_PATTERN.fullmatch(tag)
+        for name, tag in raw_versions.items()
+    ):
+        raise SyncError("versions.json 必须是 skill name 到正式 v* Release tag 的字符串映射")
+    current_versions: dict[str, str] = dict(raw_versions)
+    next_versions = dict(current_versions)
+
+    counts = {"updated": 0, "unchanged": 0, "skipped": 0}
+    staged: list[tuple[str, Path]] = []
+    with tempfile.TemporaryDirectory(prefix="agent-skills-dist-sync-") as temp:
+        for source in sources:
+            result, staged_dir = stage_source(
+                source,
+                server_url=server_url,
+                token=token,
+                current_versions=current_versions,
+                next_versions=next_versions,
+                temp_root=Path(temp),
+                opener=opener,
+                allowed_origin=allowed_origin,
+            )
+            counts[result] += 1
+            if staged_dir is not None:
+                staged.append((source["name"], staged_dir))
+
+        apply_staged_updates(staged, current_versions, next_versions, args.versions)
+
+    log(
+        "同步完成："
+        f"更新 {counts['updated']}，无变化 {counts['unchanged']}，"
+        f"无 latest Release {counts['skipped']}"
+    )
+    return 0
+
+
+if __name__ == "__main__":
+    try:
+        raise SystemExit(main())
+    except SyncError as error:
+        print(f"同步失败：{error}", file=sys.stderr)
+        raise SystemExit(1)

scripts/validate-dist.py

@@ -0,0 +1,133 @@
+#!/usr/bin/env python3
+"""Validate the distribution repository and its plugin manifest."""
+
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+from typing import Any
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+SOURCES_PATH = REPO_ROOT / ".dist-sync" / "sources.json"
+VERSIONS_PATH = REPO_ROOT / ".dist-sync" / "versions.json"
+MARKETPLACE_PATH = REPO_ROOT / ".claude-plugin" / "marketplace.json"
+
+
+class ValidationError(RuntimeError):
+    pass
+
+
+def read_json(path: Path) -> Any:
+    try:
+        return json.loads(path.read_text(encoding="utf-8"))
+    except (OSError, json.JSONDecodeError) as error:
+        raise ValidationError(f"无法读取 JSON：{path}: {error}") from error
+
+
+def frontmatter_name(skill_file: Path) -> str:
+    try:
+        lines = skill_file.read_text(encoding="utf-8-sig").splitlines()
+    except (OSError, UnicodeDecodeError) as error:
+        raise ValidationError(f"无法读取 {skill_file}: {error}") from error
+    if not lines or lines[0].strip() != "---":
+        raise ValidationError(f"SKILL.md 缺少 YAML frontmatter：{skill_file}")
+    for line in lines[1:]:
+        if line.strip() == "---":
+            break
+        if line.startswith("name:"):
+            return line.split(":", 1)[1].strip().strip("'\"")
+    raise ValidationError(f"SKILL.md frontmatter 缺少 name：{skill_file}")
+
+
+def main() -> int:
+    sources_doc = read_json(SOURCES_PATH)
+    sources = sources_doc.get("sources") if isinstance(sources_doc, dict) else None
+    if not isinstance(sources, list) or not sources:
+        raise ValidationError("sources.json 必须包含非空 sources 数组")
+
+    allowlisted: set[str] = set()
+    for source in sources:
+        if not isinstance(source, dict) or not all(
+            isinstance(source.get(field), str) and source[field] for field in ("name", "repository", "asset")
+        ):
+            raise ValidationError("sources.json 包含无效来源")
+        name = source["name"]
+        if name in allowlisted:
+            raise ValidationError(f"sources.json 包含重复名称：{name}")
+        allowlisted.add(name)
+
+    versions = read_json(VERSIONS_PATH)
+    if not isinstance(versions, dict) or not all(
+        isinstance(name, str) and isinstance(tag, str) and tag for name, tag in versions.items()
+    ):
+        raise ValidationError("versions.json 必须是非空 tag 字符串映射")
+    unknown_versions = set(versions) - allowlisted
+    if unknown_versions:
+        raise ValidationError(f"versions.json 包含未列入同步白名单的 skill：{sorted(unknown_versions)}")
+
+    marketplace = read_json(MARKETPLACE_PATH)
+    plugins = marketplace.get("plugins") if isinstance(marketplace, dict) else None
+    if not isinstance(plugins, list) or not plugins:
+        raise ValidationError("marketplace.json 必须包含非空 plugins 数组")
+
+    manifest_names: list[str] = []
+    manifest_paths: set[Path] = set()
+    for plugin in plugins:
+        skills = plugin.get("skills") if isinstance(plugin, dict) else None
+        if not isinstance(skills, list) or not skills:
+            raise ValidationError("marketplace.json 的每个 plugin 都必须列出 skills")
+        for raw_path in skills:
+            if not isinstance(raw_path, str) or not raw_path.startswith("./skills/"):
+                raise ValidationError(f"无效 marketplace skill 路径：{raw_path!r}")
+            relative = Path(raw_path[2:])
+            skill_dir = (REPO_ROOT / relative).resolve()
+            skills_root = (REPO_ROOT / "skills").resolve()
+            if skill_dir.parent != skills_root:
+                raise ValidationError(f"skill 必须直接位于 skills/<name>：{raw_path}")
+            if skill_dir in manifest_paths:
+                raise ValidationError(f"marketplace.json 重复列出路径：{raw_path}")
+            manifest_paths.add(skill_dir)
+
+            skill_file = skill_dir / "SKILL.md"
+            if not skill_file.is_file():
+                raise ValidationError(f"marketplace 路径缺少 SKILL.md：{raw_path}")
+            actual_name = frontmatter_name(skill_file)
+            if actual_name != skill_dir.name:
+                raise ValidationError(
+                    f"{raw_path}/SKILL.md 的 name 为 {actual_name!r}，目录名为 {skill_dir.name!r}"
+                )
+            manifest_names.append(actual_name)
+
+    if len(manifest_names) != len(set(manifest_names)):
+        raise ValidationError("marketplace.json 中存在重复的 skill name")
+
+    disk_dirs = {
+        path.resolve()
+        for path in (REPO_ROOT / "skills").iterdir()
+        if path.is_dir() and not path.name.startswith(".")
+    }
+    unlisted = disk_dirs - manifest_paths
+    missing = manifest_paths - disk_dirs
+    if unlisted:
+        raise ValidationError(f"skills/ 中存在未列入 marketplace 的目录：{sorted(path.name for path in unlisted)}")
+    if missing:
+        raise ValidationError(f"marketplace 指向不存在的目录：{sorted(path.name for path in missing)}")
+
+    for skill_dir in sorted(manifest_paths):
+        for path in skill_dir.rglob("*"):
+            if not path.is_file() or path.stat().st_size > 1024:
+                continue
+            if path.read_bytes().startswith(b"version https://git-lfs.github.com/spec/v1"):
+                raise ValidationError(f"发现 Git LFS pointer，分发副本缺少真实文件：{path.relative_to(REPO_ROOT)}")
+
+    print(f"分发仓校验通过：{len(manifest_names)} 个 skill，{len(allowlisted)} 个 Release 同步来源")
+    return 0
+
+
+if __name__ == "__main__":
+    try:
+        raise SystemExit(main())
+    except ValidationError as error:
+        print(f"校验失败：{error}", file=sys.stderr)
+        raise SystemExit(1)

skills/think-twice/SKILL.md

@@ -39,9 +39,9 @@ they can open right now. **Background** is what they already know.
 | can open the code  | exact locations, code, schema | the design one level up; code as an optional anchor |
 | cannot             | conclusion and boundaries     | consequences and comparisons                        |
 
-The table is a default, not a verdict. A reader in the top-left cell who is
-optimising for speed may still want the design level and no code at all. When
-it is unclear, one direct question is faster than a guess.
+The table is a default, not a verdict. A reader who can open the code and knows
+the domain may still be optimising for speed and want the design level with no
+code at all. When it is unclear, one direct question is faster than a guess.
 
 Pitch one level higher than feels natural and let them ask downward. Opening
 with code or a schema is a choice that needs a reason: they asked for it, or
@@ -57,9 +57,10 @@ meaning for it.
 
 ### 1. Locate the blocker
 
-Pick the highest one: **level** (pitched at the wrong altitude) | **concept** |
-**scope** | **steps** | **token** (undefined word or dangling pointer). Then
-name which conclusion, boundary, evidence and next step this text must carry.
+The five blockers run from highest to lowest; pick the highest that applies.
+**level** (pitched at the wrong altitude), **concept**, **scope**, **steps**,
+**token** (undefined word or dangling pointer). Then name which conclusion,
+boundary, evidence and next step this text must carry.
 
 **Done when:** one blocker is named and every load-bearing slot is listed.
 
@@ -157,7 +158,9 @@ goes, and what to do next, wherever those questions matter.
 
 ### Chinese prose for a human reader
 
-Also load [style.md](style.md), the only conditional part of this skill.
+When the output is Chinese prose a person will read, and it is not code-heavy,
+also load [style.md](style.md) and sweep its list before delivery. It is the
+only conditional part of this skill.
 
 ## When the rewrite misses
 

skills/think-twice/style.md

@@ -1,12 +1,8 @@
 # Style — 中文表述 (optional part)
 
-Load only when the output is Chinese prose for a human reader; even then, skip
-code-heavy text. This file is a guardrail against model-flavoured prose; when
-model writing no longer needs it, delete the file and every pointer to it (grep
-both skills for style.md) — neither core depends on it.
-
-Every rule is countable. Before delivery, sweep the list once: each rule passes,
-or the text changes until it does.
+A guardrail against model-flavoured prose. Every rule below names what to look
+for and what to change it to. Before delivery, sweep the list once: each rule
+passes, or the text changes until it does.
 
 ## 句子