@remixmate/cli 0.1.0

package/skills/template-bind/version.json

@@ -0,0 +1,7 @@
+{
+  "skillName": "template-bind",
+  "repoName": "agent-skill-media-maker",
+  "skillId": "475",
+  "version": "V10",
+  "skillDescription": "视频模板仓库与绑定技能，存储所有视频模板定义，根据指定模板 ID 生成 TemplateBinding。\n\n当用户提到以下任何需求时，立即使用本 skill：\n- 查看可用模板、列出所有模板\n- 为 DSL 绑定模板（需指定 template-id）\n\n当 Agent 拿到 Video DSL 需要生成 TemplateBinding 时，也应使用本 skill。"
+}

package/skills/template-bind/video_dsl/README.md

@@ -0,0 +1,97 @@
+# Video DSL 体系
+
+本目录包含视频内容生产的核心数据定义：DSL Schema、模板注册表、运行时工具和可复用组件库。
+
+## 目录结构
+
+```
+video_dsl/
+├── schema/                    # DSL Schema 定义
+│   ├── video-dsl-v1alpha1.json          # Video DSL JSON Schema
+│   ├── template-definition-v1alpha1.json # 模板定义 Schema
+│   ├── template-binding-v1alpha1.json    # 模板绑定 Schema
+│   ├── render-plan-v1alpha1.json         # RenderPlan Schema
+│   └── examples/                         # 各模板的 DSL + Binding 示例
+│       ├── ai-learning-tips.dsl.json
+│       ├── ai-learning-tips.binding.json
+│       ├── html-slide-demo.dsl.json
+│       └── ...
+├── runtime/                   # DSL 运行时工具
+│   ├── dsl_validator.py       # DSL Schema 校验
+│   ├── template_binder.py     # Scene → Slot 绑定（从 @ab-templates/metadata 读取模板）
+│   ├── timeline_compiler.py   # 时间线编译（帧号、字幕切片、旁白分段）
+│   ├── asset_adapter.py       # 素材适配器
+│   └── prompt_enhancer.py     # Prompt 增强
+└── library/                   # 可复用组件库（预留）
+    ├── blocks/                # 功能块（如 CTA、数字人、全屏背景）
+    ├── components/            # 基础组件（音频、文本、背景、装饰）
+    ├── scenes/                # 场景模板（开场钩子、知识点、绘本页）
+    └── videos/                # 完整视频模板（英文绘本等）
+```
+
+> **注意**：模板元数据（template.json / registry.json）已迁移至独立的 `template-library` 仓库，
+> 通过 `@ab-templates/metadata` 包发布。`template_binder.py` 直接读取该包产出的 `registry.json`。
+
+## 核心概念
+
+### 三个中间态
+
+| 中间态 | Schema | 职责 |
+|--------|--------|------|
+| **Video DSL** | `video-dsl-v1alpha1.json` | 描述视频应该长什么样（场景、素材、旁白） |
+| **TemplateBinding** | `template-binding-v1alpha1.json` | 描述 scene 如何映射到模板 slot |
+| **RenderPlan** | `render-plan-v1alpha1.json` | DSL 与渲染引擎之间的防腐层 |
+
+### 数据流
+
+```
+主题 → [gen-script] → Video DSL
+                          ↓
+                   [template-bind] → TemplateBinding
+                          ↓
+                   [render-video] → RenderPlan → MP4 / 剪映草稿
+```
+
+## 模板列表
+
+| 模板 ID | 名称 | 适用场景 |
+|---------|------|----------|
+| `image-slide` | 图片幻灯片 | 通用知识分享、产品介绍（含 tech 科技风变体） |
+| `picture-book-en` | 英文绘本视频 | 儿童英语启蒙、双语绘本故事 |
+| `image-slideshow` | 图片轮播视频 | 知识分享、种草笔记、技巧总结 |
+| `knowledge-slides` | 知识幻灯片 | 方法论拆解、教程讲解、知识科普 |
+| `html-slide` | 知识看板 | 知识科普、概念讲解、白板风格内容 |
+
+## runtime 模块说明
+
+| 模块 | 职责 | 被谁调用 |
+|------|------|----------|
+| `dsl_validator.py` | DSL JSON Schema 校验与参数标准化 | gen-script、render-video |
+| `template_binder.py` | 将 DSL scene 映射到模板 slot，生成 TemplateBinding（从 @ab-templates/metadata 读取模板定义） | template-bind |
+| `timeline_compiler.py` | 帧号计算、字幕切片（`split_subtitle`）、旁白分段（`segment_narration`） | render-video（跨 skill import） |
+| `asset_adapter.py` | 素材格式适配 | render-video |
+| `prompt_enhancer.py` | 图片 prompt 增强 | render-video |
+
+**注意**：`timeline_compiler.py` 被 `render-video/scripts/render_video.py` 直接 import，修改函数签名需两个 skill 同步测试。
+
+## library 组件库（预留）
+
+`library/` 目录采用四层分级设计，用于存放可复用的 DSL 片段：
+
+- **components/**：最小粒度的基础组件（音频、文本、背景、头像、装饰、媒体、叠加层）
+- **blocks/**：由多个 component 组合而成的功能块（CTA 按钮、数字人区域、全屏背景图、标题区、产品卡片、配音条、双语字幕）
+- **scenes/**：完整的场景模板（开场钩子、知识点讲解、绘本封面/内页/结尾、CTA 结尾、数字人口播）
+- **videos/**：完整的视频模板（英文绘本等）
+
+组件库当前为目录骨架，后续可填充具体的 JSON 片段供 Agent 和 gen_script.py 复用。
+
+## 示例文件
+
+`schema/examples/` 下包含各模板的完整 DSL + Binding 示例，是 Agent 生成 DSL 时的权威参考：
+
+- `ai-learning-tips.*` — image-slide 模板示例
+- `html-slide-demo.*` — html-slide 模板示例
+- `knowledge-slides-demo.*` — knowledge-slides 模板示例
+- `animal-friends-picturebook.*` — picture-book-en 模板示例
+
+Agent 在为指定模板生成 DSL 时，必须先读取对应示例，按模板原生要求生成。

package/skills/template-bind/video_dsl/runtime/__init__.py

@@ -0,0 +1,11 @@
+"""Package marker for ``video_dsl.runtime`` — kept intentionally empty.
+
+This file exists so ``from video_dsl.runtime.<mod> import …`` resolves when
+``skills/template-bind/`` is on sys.path (the pattern used by render-video
+and gen-script). Public modules:
+
+    - timeline_compiler  : DSL → Remotion frame-axis timeline + subtitle splits
+    - prompt_enhancer    : LLM image-prompt enrichment used by gen-script
+    - template_binder    : registry summary writer (sync_registry helper)
+    - dsl_validator      : DSL validation (structural / integrity / full presets)
+"""

package/skills/template-bind/video_dsl/runtime/dsl_validator.py

@@ -0,0 +1,495 @@
+"""
+DSL 校验器 - 对 Video DSL v1alpha1 做结构校验与标准化。
+
+模块结构（T6 重构后）：
+
+* 细粒度 ``_check_*`` 检查器：每个函数检查一个独立维度，返回
+  ``list[ValidationError]``。
+* 三个 preset 聚合器，对应历史上 gen-script / render-video / 完整版三种
+  使用场景：
+    - ``validate_structural(dsl)``  ⇐ gen-script 历史调用点
+    - ``validate_integrity(dsl)``   ⇐ render-video 历史调用点
+    - ``validate_dsl(dsl)``         ⇐ 完整版（含 capabilities-driven 语言约束、字段枚举等）
+  历史的两个 caller 现在 thin-wrap 对应 preset，**所检规则与历史完全等价**，
+  以保证不引入任何"原本能过、现在被拒"的行为变化。
+
+模块以前同时存在三份算法（在 gen_script.py、render_video.py 与本文件内），
+``lineSyncSlides`` 注册表查询、items 数对齐、模板专属语言约束等代码在三份
+之间已经静默漂移。统一到本文件后，所有维度只有一份实现。所有模板差异通过
+``template.json`` 的 ``capabilities`` 字段声明，本文件读取并应用 —— 新增需要特殊
+处理的模板时**不需要再改 ab-skill 代码**。
+"""
+
+import re
+from typing import Iterable, Optional
+
+VALID_VERSIONS = {"v1alpha1"}
+VALID_PLATFORMS = {"douyin", "xiaohongshu", "bilibili", "wechat", "youtube", "generic"}
+VALID_RATIOS = {"16:9", "9:16", "1:1", "4:3", "3:4", "21:9"}
+VALID_RESOLUTIONS = {"480p", "720p", "1080p", "4k"}
+VALID_PURPOSES = {"opening", "intro", "point", "example", "explanation", "transition", "highlight", "cta", "ending"}
+VALID_LAYOUTS = {"full-visual", "split-left-right", "split-top-bottom", "picture-in-picture", "text-overlay", "avatar-with-bg", "kenburns"}
+VALID_ASSET_TYPES = {"image", "video", "audio", "avatar", "subtitle", "bgm"}
+VALID_ASSET_SOURCES = {"existing", "gen-image", "gen-video", "gen-voice", "gen-digital-human"}
+
+# CJK 汉字范围（简繁通用）+ 常见中文标点
+_CJK_PATTERN = re.compile(r"[一-鿿㐀-䶿＀-￯　-〿]")
+
+
+# ════════════════════════════════════════════════════════════════════════════
+# Error type
+# ════════════════════════════════════════════════════════════════════════════
+
+
+class ValidationError:
+    def __init__(self, path: str, message: str, severity: str = "error"):
+        self.path = path
+        self.message = message
+        self.severity = severity
+
+    def __str__(self) -> str:
+        return f"[{self.severity.upper()}] {self.path}: {self.message}"
+
+    def to_dict(self) -> dict:
+        return {"path": self.path, "message": self.message, "severity": self.severity}
+
+
+def errors_as_strings(errors: Iterable[ValidationError]) -> list[str]:
+    """Convert ValidationError objects to ``"path: message"`` strings.
+
+    Designed for the two historical callers (gen_script / render_video) that
+    print ``f"   - {err}"`` after a header line. We drop the ``[ERROR]`` /
+    ``[WARNING]`` prefix here because the prior callers' output didn't have
+    it; the call site already announced "validation failed".
+    """
+    return [f"{e.path}: {e.message}" for e in errors]
+
+
+# ════════════════════════════════════════════════════════════════════════════
+# Helpers
+# ════════════════════════════════════════════════════════════════════════════
+
+
+def _pick_template_id(dsl: dict) -> Optional[str]:
+    """Pick the templateId from DSL meta (single source of truth).
+
+    ``renderHints.templatePreference`` was the legacy location and is no
+    longer generated by gen_script. We still tolerate it during a
+    transition period, but new authors should write ``meta.templateId``.
+    """
+    meta = dsl.get("meta") or {}
+    if isinstance(meta, dict):
+        tid = meta.get("templateId")
+        if isinstance(tid, str) and tid:
+            return tid
+    hints = dsl.get("renderHints") or {}
+    prefs = hints.get("templatePreference") or []
+    if isinstance(prefs, list) and prefs:
+        return prefs[0]
+    return hints.get("templateId") if isinstance(hints, dict) else None
+
+
+def _line_sync_slides_for_template(template_id: Optional[str]) -> dict[str, str]:
+    """Look up ``lineSyncSlides`` for a template from registry.json.
+
+    新增多项卡片幻灯片只需在 template.json 加一行 ``lineSyncSlides[slideId] = key``，
+    不再需要改 ab-skill。registry 读取失败（环境缺 registry_loader、HTTP 不通等）
+    时返回空 dict，调用方将跳过相关校验。
+
+    O(1) lookup via ``registry_loader.get_template``; safe to call once per
+    scene at 100+ templates without scanning the full list.
+    """
+    tpl = _get_template_cfg(template_id)
+    if not tpl:
+        return {}
+    spec = tpl.get("lineSyncSlides") or {}
+    return {k: v for k, v in spec.items() if isinstance(v, str)}
+
+
+def _get_template_cfg(template_id: Optional[str]) -> Optional[dict]:
+    """Indexed registry lookup with the legacy import fallback.
+
+    Returns None silently when:
+      - ``template_id`` is empty
+      - ``registry_loader`` cannot be imported (e.g. unit tests outside the
+        shared-lib sys.path)
+      - the registry can't be loaded (network/file/etc.)
+      - the template isn't in the visible set (e.g. beta gate hides it)
+    """
+    if not template_id:
+        return None
+    try:
+        from registry_loader import get_template  # type: ignore
+    except ImportError:
+        return None
+    try:
+        return get_template(template_id)
+    except Exception:
+        return None
+
+
+def _contains_cjk(text: str) -> bool:
+    return bool(text) and bool(_CJK_PATTERN.search(text))
+
+
+# ════════════════════════════════════════════════════════════════════════════
+# Granular checks
+# ════════════════════════════════════════════════════════════════════════════
+
+
+def _check_version(dsl: dict) -> list[ValidationError]:
+    if dsl.get("version") not in VALID_VERSIONS:
+        return [ValidationError(
+            "version",
+            f"must be one of {sorted(VALID_VERSIONS)}, got '{dsl.get('version')}'",
+        )]
+    return []
+
+
+def _check_meta_presence(dsl: dict) -> list[ValidationError]:
+    """gen-script 历史:meta 必须存在且至少有 title。"""
+    errors: list[ValidationError] = []
+    meta = dsl.get("meta")
+    if not isinstance(meta, dict):
+        errors.append(ValidationError("meta", "required and must be an object"))
+        return errors
+    if not meta.get("title"):
+        errors.append(ValidationError("meta.title", "required"))
+    return errors
+
+
+def _check_meta_enums(dsl: dict) -> list[ValidationError]:
+    """完整版额外:meta.platform 必须是已知枚举。"""
+    errors: list[ValidationError] = []
+    meta = dsl.get("meta")
+    if isinstance(meta, dict):
+        platform = meta.get("platform", "generic")
+        if platform not in VALID_PLATFORMS:
+            errors.append(ValidationError(
+                "meta.platform", f"must be one of {sorted(VALID_PLATFORMS)}",
+            ))
+    return errors
+
+
+def _check_global_presence(dsl: dict) -> list[ValidationError]:
+    """gen-script 历史:global 必须存在(只检查存在性)。"""
+    if not isinstance(dsl.get("global"), dict):
+        return [ValidationError("global", "required and must be an object")]
+    return []
+
+
+def _check_global_enums(dsl: dict) -> list[ValidationError]:
+    """完整版额外:global.aspectRatio / resolution 必须是已知枚举。"""
+    errors: list[ValidationError] = []
+    g = dsl.get("global")
+    if isinstance(g, dict):
+        ratio = g.get("aspectRatio", "16:9")
+        if ratio not in VALID_RATIOS:
+            errors.append(ValidationError(
+                "global.aspectRatio", f"must be one of {sorted(VALID_RATIOS)}",
+            ))
+        res = g.get("resolution", "1080p")
+        if res not in VALID_RESOLUTIONS:
+            errors.append(ValidationError(
+                "global.resolution", f"must be one of {sorted(VALID_RESOLUTIONS)}",
+            ))
+    return errors
+
+
+def _check_scenes_min(dsl: dict) -> list[ValidationError]:
+    """Scenes 至少 1。两个 preset 共用。"""
+    scenes = dsl.get("scenes", [])
+    if not isinstance(scenes, list) or len(scenes) == 0:
+        return [ValidationError("scenes", "must contain at least 1 scene")]
+    return []
+
+
+def _check_scenes_max(dsl: dict) -> list[ValidationError]:
+    """Scenes 上限 20。仅 gen-script 历史规则集 + 完整版用——render-video
+    历史不查上限,合并时若同时启用此检查会"原本能过的 DSL 现在被拒",
+    故只在 ``validate_structural`` / ``validate_dsl`` 启用。
+    """
+    scenes = dsl.get("scenes", [])
+    if isinstance(scenes, list) and len(scenes) > 20:
+        return [ValidationError(
+            "scenes", f"must contain at most 20 scenes, got {len(scenes)}",
+        )]
+    return []
+
+
+def _check_scene_required_fields(dsl: dict) -> list[ValidationError]:
+    """gen-script 历史:每个 scene 必须有 id 和 purpose。"""
+    errors: list[ValidationError] = []
+    for i, scene in enumerate(dsl.get("scenes", []) or []):
+        if not scene.get("id"):
+            errors.append(ValidationError(f"scenes[{i}].id", "required"))
+        if not scene.get("purpose"):
+            errors.append(ValidationError(f"scenes[{i}].purpose", "required"))
+    return errors
+
+
+def _check_scene_id_uniqueness(dsl: dict) -> list[ValidationError]:
+    """完整版额外:scene.id 唯一性。"""
+    errors: list[ValidationError] = []
+    seen: set = set()
+    for i, scene in enumerate(dsl.get("scenes", []) or []):
+        sid = scene.get("id")
+        if sid and sid in seen:
+            errors.append(ValidationError(
+                f"scenes[{i}].id", f"duplicate scene id '{sid}'",
+            ))
+        elif sid:
+            seen.add(sid)
+    return errors
+
+
+def _check_scene_enum_warnings(dsl: dict) -> list[ValidationError]:
+    """完整版额外:scene.purpose / scene.layout 枚举(severity=warning)。"""
+    errors: list[ValidationError] = []
+    for i, scene in enumerate(dsl.get("scenes", []) or []):
+        purpose = scene.get("purpose")
+        if purpose and purpose not in VALID_PURPOSES:
+            errors.append(ValidationError(
+                f"scenes[{i}].purpose",
+                f"must be one of {sorted(VALID_PURPOSES)}", "warning",
+            ))
+        layout = scene.get("layout")
+        if layout and layout not in VALID_LAYOUTS:
+            errors.append(ValidationError(
+                f"scenes[{i}].layout",
+                f"must be one of {sorted(VALID_LAYOUTS)}", "warning",
+            ))
+    return errors
+
+
+def _check_asset_duplicates(dsl: dict) -> list[ValidationError]:
+    """render-video 历史:assets 数组里 assetId 重复检查。"""
+    errors: list[ValidationError] = []
+    counts: dict[str, int] = {}
+    for a in dsl.get("assets", []) or []:
+        aid = a.get("assetId", "")
+        counts[aid] = counts.get(aid, 0) + 1
+    for aid, cnt in counts.items():
+        if cnt > 1:
+            errors.append(ValidationError(
+                f"assets[{aid}]", f"duplicate assetId '{aid}' (appears {cnt} times)",
+            ))
+    return errors
+
+
+def _check_asset_enums(dsl: dict) -> list[ValidationError]:
+    """完整版额外:asset.type / asset.source 枚举。"""
+    errors: list[ValidationError] = []
+    for i, asset in enumerate(dsl.get("assets", []) or []):
+        atype = asset.get("type")
+        if atype and atype not in VALID_ASSET_TYPES:
+            errors.append(ValidationError(
+                f"assets[{i}].type", f"must be one of {sorted(VALID_ASSET_TYPES)}",
+            ))
+        asource = asset.get("source")
+        if asource and asource not in VALID_ASSET_SOURCES:
+            errors.append(ValidationError(
+                f"assets[{i}].source", f"must be one of {sorted(VALID_ASSET_SOURCES)}",
+            ))
+    return errors
+
+
+def _check_scene_asset_refs(dsl: dict) -> list[ValidationError]:
+    """render-video 历史:scene 引用的 assetRef 必须在 assets[] 声明过。
+
+    检查范围:
+      - scene.audio.narration.assetRef
+      - scene.visuals.background.assetRef
+    """
+    errors: list[ValidationError] = []
+    declared = {
+        a.get("assetId") for a in (dsl.get("assets") or []) if a.get("assetId")
+    }
+    for idx, scene in enumerate(dsl.get("scenes", []) or []):
+        sid = scene.get("id", f"scenes[{idx}]")
+        nar = (scene.get("audio") or {}).get("narration") or {}
+        nar_ref = nar.get("assetRef")
+        if nar_ref and nar_ref not in declared:
+            errors.append(ValidationError(
+                f"scenes[{idx}].audio.narration.assetRef",
+                f"scene '{sid}': narration.assetRef '{nar_ref}' not found in assets[]",
+            ))
+        bg_ref = ((scene.get("visuals") or {}).get("background") or {}).get("assetRef")
+        if bg_ref and bg_ref not in declared:
+            errors.append(ValidationError(
+                f"scenes[{idx}].visuals.background.assetRef",
+                f"scene '{sid}': visuals.background.assetRef '{bg_ref}' not found in assets[]",
+            ))
+    return errors
+
+
+def _check_assetbindings_deprecation(dsl: dict) -> list[ValidationError]:
+    """完整版额外:assetBindings 引用未知 id 时给 warning + 提示已弃用。"""
+    errors: list[ValidationError] = []
+    declared = {
+        a.get("assetId") for a in (dsl.get("assets") or []) if a.get("assetId")
+    }
+    for i, scene in enumerate(dsl.get("scenes", []) or []):
+        for ref in scene.get("assetBindings", []) or []:
+            if ref not in declared:
+                errors.append(ValidationError(
+                    f"scenes[{i}].assetBindings",
+                    f"references unknown assetId '{ref}' (note: assetBindings is deprecated, "
+                    "use visuals.background.assetRef + audio.narration.assetRef instead)",
+                    "warning",
+                ))
+    return errors
+
+
+def _check_narration_items_count(dsl: dict) -> list[ValidationError]:
+    """render-video / 完整版:结构化 narration items 数与 templateData 的对应数组等长。
+
+    哪些 slideId 走该校验,由模板的 ``lineSyncSlides`` 字段声明(P2.2 起);
+    registry 不可用或模板没声明则跳过。两个 preset 共用本检查器,统一来源
+    避免 render-video 与完整版的算法继续漂移。
+    """
+    errors: list[ValidationError] = []
+    template_id = _pick_template_id(dsl)
+    line_sync_slides = _line_sync_slides_for_template(template_id)
+
+    for idx, scene in enumerate(dsl.get("scenes", []) or []):
+        sid = scene.get("id", f"scenes[{idx}]")
+        narration = (scene.get("audio") or {}).get("narration") or {}
+        items = narration.get("items")
+        if not items or not isinstance(items, list):
+            continue
+        custom = scene.get("customPayload") or {}
+        slide_id = custom.get("slideId") or (custom.get("templateData") or {}).get("slideId")
+        tdata = custom.get("templateData") or scene.get("templateData") or {}
+        items_key = line_sync_slides.get(slide_id or "")
+        if not items_key:
+            continue
+        card_list = tdata.get(items_key)
+        if not isinstance(card_list, list):
+            continue
+        # render-video 历史只在 actual!=expected 时报;完整版也是同样语义。
+        actual = len([x for x in items if (isinstance(x, str) and x.strip()) or isinstance(x, dict)])
+        expected = len(card_list)
+        if actual != expected:
+            errors.append(ValidationError(
+                f"scenes[{idx}].audio.narration.items",
+                f"scene '{sid}': narration.items has {actual} entries "
+                f"but templateData.{items_key} has {expected} — counts must match "
+                "so each card lines up with one narration line",
+            ))
+    return errors
+
+
+def _check_narration_language_strict(dsl: dict) -> list[ValidationError]:
+    """完整版:模板若声明 ``capabilities.narrationLanguageStrict`` 则强约束旁白语言。
+
+    Generic capability-driven replacement for what used to be a hardcoded
+    ``picture-book-en`` CJK check. Templates declare which language the
+    narration MUST be in (currently supported: ``"en"`` rejects any CJK
+    character; ``"zh"`` is a placeholder reserved for future symmetric
+    enforcement — we don't yet flag latin-only narration on zh templates
+    because gen-script regularly mixes brand names in English).
+
+    Detection points are unchanged from the legacy check:
+      - every ``gen-voice`` asset's ``payload.text``
+      - every scene's ``audio.narration.text``
+    """
+    tpl = _get_template_cfg(_pick_template_id(dsl))
+    if not tpl:
+        return []
+    strict = (tpl.get("capabilities") or {}).get("narrationLanguageStrict")
+    if strict != "en":
+        # Today only the "narration must be English" case is enforced;
+        # adding "zh" / other languages goes here without touching callers.
+        return []
+    tid = tpl.get("templateId", "<unknown>")
+    msg = (
+        f"narration on template '{tid}' must be English only (no CJK characters); "
+        "put the Chinese translation in textLayers[role=subheadline].content instead"
+    )
+    errors: list[ValidationError] = []
+    for i, asset in enumerate(dsl.get("assets", []) or []):
+        if asset.get("source") != "gen-voice":
+            continue
+        voice_text = ((asset.get("payload") or {}).get("text")) or ""
+        if _contains_cjk(voice_text):
+            errors.append(ValidationError(f"assets[{i}].payload.text", msg))
+    for idx, scene in enumerate(dsl.get("scenes", []) or []):
+        narration = (scene.get("audio") or {}).get("narration") or {}
+        narration_text = narration.get("text")
+        if isinstance(narration_text, str) and _contains_cjk(narration_text):
+            errors.append(ValidationError(
+                f"scenes[{idx}].audio.narration.text", msg,
+            ))
+    return errors
+
+
+# ════════════════════════════════════════════════════════════════════════════
+# Aggregator presets
+# ════════════════════════════════════════════════════════════════════════════
+
+
+def validate_structural(dsl: dict) -> list[ValidationError]:
+    """gen-script 历史规则集 — 仅检查 DSL 整体结构 / 必填字段。
+
+    与历史 ``gen_script.validate_dsl`` 同集合:
+      version + meta(title) + global(presence) + scene 数 1–20 + 每场 id/purpose。
+
+    不检查:asset 一致性、enum 枚举、capabilities.narrationLanguageStrict——保持与历史等价。
+    """
+    errors: list[ValidationError] = []
+    errors += _check_version(dsl)
+    errors += _check_meta_presence(dsl)
+    errors += _check_global_presence(dsl)
+    errors += _check_scenes_min(dsl)
+    errors += _check_scenes_max(dsl)
+    errors += _check_scene_required_fields(dsl)
+    return errors
+
+
+def validate_integrity(dsl: dict) -> list[ValidationError]:
+    """render-video 历史规则集 — 关注 asset 引用一致性 + 数量对齐。
+
+    与历史 ``render_video.validate_dsl`` 同集合:
+      version + scene 数 + assetId 去重 + assetRef 可解析 + 结构化 narration items 数。
+
+    不检查:meta.title、global 必填、enum 枚举、capabilities.narrationLanguageStrict——保持等价。
+    """
+    errors: list[ValidationError] = []
+    errors += _check_version(dsl)
+    errors += _check_scenes_min(dsl)
+    # NOTE: render-video 历史只查下限,不查 ≤20 上限——合并时若同启用上限会
+    # 把"超 20 场但其它都对"的 DSL 从「通过」改成「拒绝」,违反"重构不影响功能"。
+    errors += _check_asset_duplicates(dsl)
+    errors += _check_scene_asset_refs(dsl)
+    errors += _check_narration_items_count(dsl)
+    return errors
+
+
+def validate_dsl(dsl: dict) -> list[ValidationError]:
+    """完整 DSL 校验:结构 + 一致性 + 枚举 + capabilities.narrationLanguageStrict。
+
+    历史上 dsl_validator 暴露的就是这版,无外部 caller(三个真 caller 都
+    走 ``validate_structural`` / ``validate_integrity``)。本函数面向想要"全
+    部规则"的新代码 —— 例如 CI 校验或人工 ``--validate-strict`` 入口。
+    """
+    errors: list[ValidationError] = []
+    errors += _check_version(dsl)
+    errors += _check_meta_presence(dsl)
+    errors += _check_meta_enums(dsl)
+    errors += _check_global_presence(dsl)
+    errors += _check_global_enums(dsl)
+    errors += _check_scenes_min(dsl)
+    errors += _check_scenes_max(dsl)
+    errors += _check_scene_required_fields(dsl)
+    errors += _check_scene_id_uniqueness(dsl)
+    errors += _check_scene_enum_warnings(dsl)
+    errors += _check_asset_duplicates(dsl)
+    errors += _check_asset_enums(dsl)
+    errors += _check_scene_asset_refs(dsl)
+    errors += _check_assetbindings_deprecation(dsl)
+    errors += _check_narration_items_count(dsl)
+    errors += _check_narration_language_strict(dsl)
+    return errors
+
+