@remixmate/cli 0.9.47 → 0.9.48

package/dist/manifest.json

@@ -1,7 +1,7 @@
 {
   "schemaVersion": 1,
-  "version": "0.9.47",
-  "generatedAt": "2026-09-24T01:36:07.207Z",
+  "version": "0.9.48",
+  "generatedAt": "2026-09-24T03:37:24.431Z",
   "skills": [
     {
       "id": "doc-parse",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@remixmate/cli",
-  "version": "0.9.47",
+  "version": "0.9.48",
   "description": "AI media generation skills for Claude Code / Codex — 12 skills covering image, video, voice, digital human, web screenshot, web recording, script, template registry, rendering, Jianying export, and video deconstruction.",
   "type": "module",
   "main": "dist/index.js",
@@ -41,7 +41,8 @@
     "test:prepare-contracts": "PYTHONDONTWRITEBYTECODE=1 python3 scripts/test-prepare-contracts.py",
     "test:image-sequence": "PYTHONDONTWRITEBYTECODE=1 python3 scripts/test-image-sequence.py",
     "test:ranking-count": "PYTHONDONTWRITEBYTECODE=1 python3 scripts/test-ranking-count.py",
-    "test:image-prompt-compose": "PYTHONDONTWRITEBYTECODE=1 python3 scripts/test-image-prompt-compose.py"
+    "test:image-prompt-compose": "PYTHONDONTWRITEBYTECODE=1 python3 scripts/test-image-prompt-compose.py",
+    "test:picture-book-contract": "PYTHONDONTWRITEBYTECODE=1 python3 scripts/test-picture-book-contract.py"
   },
   "keywords": [
     "claude-code",

package/skills/gen-script/scripts/gen_script.py

@@ -628,9 +628,58 @@ def _resolve_contract(template_config: dict | None) -> dict:
         "primary_visual_type": primary_visual_type,
         "default_layout": caps.get("defaultLayout"),
         "authored_scenes": caps.get("authoredScenes") is True,
+        # 画面文字契约（dsl_validator 按 error 执行）；骨架据此为每个必填 role 留占位
+        "text_layer_rules": caps.get("textLayerRules") if isinstance(caps.get("textLayerRules"), dict) else None,
+        # arc 叙事弧的收尾场景用途（缺省 cta）。绘本收尾是 ending（寓意页），不是「关注我们」
+        "arc_closing_purpose": caps.get("arcClosingPurpose") if isinstance(caps.get("arcClosingPurpose"), str) else None,
+        # 骨架里 transitions.default 的取值；缺省 fade（历史行为）
+        "default_transition": caps.get("defaultTransition") if isinstance(caps.get("defaultTransition"), str) else None,
     }
 
 
+def _slot_reads_visual(template_config: dict | None, purpose: str) -> bool:
+    """这个用途的版式会不会显示背景素材（slotMapping 里有没有读 visuals.background.assetRef）。
+
+    不显示的版式（如绘本的 ending 寓意页）不该规划图片：素材照样会被生成、扣积分，
+    却永远不上屏。模板没声明 slotMapping 时按「会显示」处理，保持历史行为。
+    """
+    mapping = (template_config or {}).get("slotMapping")
+    if not isinstance(mapping, dict) or not mapping:
+        return True
+    slot = mapping.get(purpose) or mapping.get("default") or {}
+    seen: set = set()
+    while isinstance(slot, dict) and "$ref" in slot and slot["$ref"] not in seen:
+        seen.add(slot["$ref"])
+        slot = mapping.get(slot["$ref"]) or {}
+    extractors = (slot or {}).get("propExtractors") if isinstance(slot, dict) else None
+    if not isinstance(extractors, dict):
+        return True
+    return any(
+        isinstance(ex, dict) and str(ex.get("from", "")).endswith("visuals.background.assetRef")
+        for ex in extractors.values()
+    )
+
+
+def _placeholder_text_layers(rules: dict) -> list[dict]:
+    """模板声明了 capabilities.textLayerRules 时，骨架的文字层 = 每个必填 role 一条占位。
+
+    通用骨架给的是「内容要点 N」「关注我们」这类文案，对有文字契约的模板全是错的，
+    而且留着一个没声明的 role（badge）只会让模型以为那里要填东西。占位带骨架前缀，
+    dsl_validator 会拒绝原样提交；带上语言要求，模型不用再去 llmHint 里找。
+    """
+    language = rules.get("language") if isinstance(rules.get("language"), dict) else {}
+    hints = {"en": "English text for this page", "zh": "本页中文译文"}
+    layers = []
+    for role in rules.get("required") or []:
+        if not isinstance(role, str):
+            continue
+        lang = language.get(role)
+        prefix = SKELETON_PREFIXES[1] if lang == "en" else SKELETON_PREFIXES[0]
+        hint = hints.get(lang, role)
+        layers.append({"role": role, "content": f"{prefix} {role}: {hint}"})
+    return layers
+
+
 def _normalize_linebreaks(value: str | None) -> str:
     """把画面文字里的**字面量** ``\\n`` 折成真换行。
 
@@ -794,11 +843,15 @@ def _plan_contract_scenes(
     这类模板由此能产出真正的单场景，而不再被强行补到 3 段。``--scenes`` 只对 arc 生效。
     """
     if contract["scene_strategy"] == "arc":
-        return plan_scenes(
+        plans = plan_scenes(
             topic, duration, scene_count, False,
             language=language,
             max_scenes=contract.get("max_scenes", DEFAULT_MAX_SCENES),
         )
+        closing = contract.get("arc_closing_purpose")
+        if closing and plans and plans[-1]["purpose"] == "cta":
+            plans[-1] = {**plans[-1], "purpose": closing}
+        return plans
 
     fixed = contract["fixed_scenes"] or [{"purpose": "opening"}]
     n = len(fixed)
@@ -1482,8 +1535,11 @@ def build_dsl(
             if needs_narration else ""
         )
 
+        # 本场景的版式不显示背景素材（如绘本 ending）→ 不规划视觉素材，免得白生成
+        scene_has_visual = has_visual and _slot_reads_visual(template_config, plan["purpose"])
+
         # 仅 visual-overlay 模板生成视觉素材（slide 无视觉素材，carousel-caption 已提前返回）
-        if has_visual and primary_visual_type == "image":
+        if scene_has_visual and primary_visual_type == "image":
             slot = provided_slots[idx] if idx < len(provided_slots) else ""
             slot_is_ai, slot_prompt = _parse_ai_slot(slot)
             if slot and not slot_is_ai:
@@ -1540,7 +1596,7 @@ def build_dsl(
                     "status": "planned",
                     "payload": img_payload,
                 })
-        elif has_visual and primary_visual_type == "video":
+        elif scene_has_visual and primary_visual_type == "video":
             if stub_video_url:
                 assets.append({
                     "assetId": visual_asset_id,
@@ -1587,7 +1643,7 @@ def build_dsl(
                 },
             })
 
-        if has_visual:
+        if scene_has_visual:
             layout = "text-overlay" if plan["purpose"] in ("opening", "cta") else "full-visual"
             text_layers = []
             if plan["purpose"] == "opening":
@@ -1612,6 +1668,8 @@ def build_dsl(
                 ]
                 if effective_subheadline:
                     text_layers.append({"role": "subheadline", "content": effective_subheadline, "animation": "fade-in"})
+            if contract["text_layer_rules"]:
+                text_layers = _placeholder_text_layers(contract["text_layer_rules"])
 
             scene = {
                 "id": scene_id,
@@ -1679,6 +1737,9 @@ def build_dsl(
                 if effective_subheadline:
                     text_layers.append({"role": "subheadline", "content": effective_subheadline, "animation": "fade-in"})
 
+            if contract["text_layer_rules"]:
+                text_layers = _placeholder_text_layers(contract["text_layer_rules"])
+
             custom_payload = _build_custom_payload(template_config)
             if beat_count is not None:
                 # Field names come from the template's countdown declaration — this
@@ -1737,7 +1798,8 @@ def build_dsl(
         },
         "assets": assets,
         "scenes": scenes,
-        "transitions": {"default": "fade", "duration": 0.5},
+        # 模板声明了 capabilities.defaultTransition 就用它（绘本 = none：淡入会露出黑底）
+        "transitions": {"default": contract["default_transition"] or "fade", "duration": 0.5},
     }
 
     if template_id:

package/skills/render-video/scripts/render_video.py

@@ -1440,6 +1440,26 @@ def _auto_highlight_map(entry: dict, narration_asset: Optional[dict], render_pla
     })
 
 
+def _template_default_tail_pad_frames(template_id: str, fps: int) -> int | None:
+    """模板声明的旁白尾巴默认值（``capabilities.defaultTailPadSec``），换算成帧。
+
+    场景自己的 ``customPayload.tailPadSec`` 仍然优先；这里只替换全局 1.5s 兜底。
+    为什么要模板级：逐场景设置靠模型照着 llmHint 抄，实际经常漏 —— 2026-09-24
+    线上《狼来了》绘本每页结尾 1.9s 空白，86s 里 15s 没有声音。
+    """
+    if not template_id:
+        return None
+    try:
+        from registry_loader import get_template  # type: ignore
+        tpl = get_template(template_id, include_all_statuses=True)
+    except Exception:  # noqa: BLE001 - 取不到模板就走全局默认，不因为这个挡渲染
+        return None
+    value = ((tpl or {}).get("capabilities") or {}).get("defaultTailPadSec")
+    if isinstance(value, bool) or not isinstance(value, (int, float)) or value < 0:
+        return None
+    return int(round(float(value) * fps))
+
+
 def adjust_timeline_to_audio(render_plan: dict) -> dict:
     """Post-process: use real TTS audio durations and timestamps to fix
     scene durations and subtitle segments."""
@@ -1449,6 +1469,9 @@ def adjust_timeline_to_audio(render_plan: dict) -> dict:
     assets_map = {a["assetId"]: a for a in render_plan["resolvedAssets"]}
     current_frame = 0
     adjusted = 0
+    default_tail_pad = _template_default_tail_pad_frames(render_plan.get("templateId", ""), fps)
+    if default_tail_pad is None:
+        default_tail_pad = int(fps * 1.5)
 
     for entry in render_plan["timeline"]:
         # narration assetId 现在直接由 propExtractors 注入到 props.narrationAssetId，
@@ -1472,13 +1495,14 @@ def adjust_timeline_to_audio(render_plan: dict) -> dict:
                 })
             else:
                 audio_frames = math.ceil(audio_ms / 1000.0 * fps)
-                # DSL 里 scene.customPayload.tailPadSec 可覆盖默认 1.5s 尾巴
-                # （比如开场卡片要紧接下一场，设 0 直接贴音频结束）。
+                # DSL 里 scene.customPayload.tailPadSec 可覆盖默认尾巴
+                # （比如开场卡片要紧接下一场，设 0 直接贴音频结束）；默认值是模板的
+                # capabilities.defaultTailPadSec，没声明时 1.5s。
                 tail_pad_override = entry.get("tailPadFrames")
                 tail_pad = (
                     int(tail_pad_override)
                     if isinstance(tail_pad_override, int) and tail_pad_override >= 0
-                    else int(fps * 1.5)
+                    else default_tail_pad
                 )
                 min_dur = int(fps * 2)     # minimum 2 seconds per scene
                 # 用户在 DSL 里指定的 scene 下限（minDurationSec → minDurationFrames），

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

@@ -1132,6 +1132,77 @@ def _check_narration_language_strict(dsl: dict) -> list[ValidationError]:
     return errors
 
 
+_TEXT_LAYER_SKELETON_PREFIXES = ("[骨架待填充]", "[skeleton placeholder]")
+
+
+def _check_text_layer_rules(dsl: dict) -> list[ValidationError]:
+    """模板声明 ``capabilities.textLayerRules`` 时，逐场景检查画面文字层。
+
+    声明形状::
+
+        "textLayerRules": {
+          "required": ["headline", "subheadline"],          # 每个场景都必须有且非空
+          "language": {"headline": "en", "subheadline": "zh"}  # en = 不含中文；zh = 须含中文
+        }
+
+    为什么是 error 而不是 llmHint：2026-09-24 线上《狼来了》绘本整本没有一行中文译文，
+    封面把中英文都塞进了 headline —— llmHint 写得清清楚楚，模型没照做，而当时没有
+    任何一道门拦它，成片照出，书页上留着一大块给译文的空白。
+    """
+    tpl = _get_template_cfg(_pick_template_id(dsl))
+    if not tpl:
+        return []
+    rules = (tpl.get("capabilities") or {}).get("textLayerRules")
+    if not isinstance(rules, dict):
+        return []
+    required = [r for r in (rules.get("required") or []) if isinstance(r, str)]
+    language = rules.get("language") if isinstance(rules.get("language"), dict) else {}
+    tid = tpl.get("templateId", "<unknown>")
+
+    errors: list[ValidationError] = []
+    for idx, scene in enumerate(dsl.get("scenes", []) or []):
+        sid = scene.get("id", f"#{idx}")
+        layers: dict[str, tuple[int, str]] = {}
+        for li, layer in enumerate(scene.get("textLayers") or []):
+            if not isinstance(layer, dict):
+                continue
+            role = layer.get("role")
+            content = layer.get("content")
+            if isinstance(role, str) and role not in layers:
+                layers[role] = (li, content.strip() if isinstance(content, str) else "")
+
+        for role in required:
+            li, content = layers.get(role, (None, ""))
+            path = f"scenes[{idx}].textLayers" + (f"[{li}].content" if li is not None else "")
+            if not content:
+                errors.append(ValidationError(
+                    path,
+                    f"scene '{sid}': template '{tid}' requires a non-empty textLayers role '{role}'"
+                    + (f" (language: {language[role]})" if role in language else ""),
+                ))
+            elif content.startswith(_TEXT_LAYER_SKELETON_PREFIXES):
+                errors.append(ValidationError(
+                    path, f"scene '{sid}': textLayers role '{role}' is still the gen_script placeholder",
+                ))
+
+        for role, lang in language.items():
+            li, content = layers.get(role, (None, ""))
+            if li is None or not content or content.startswith(_TEXT_LAYER_SKELETON_PREFIXES):
+                continue
+            path = f"scenes[{idx}].textLayers[{li}].content"
+            if lang == "en" and _contains_cjk(content):
+                others = [r for r, l in language.items() if l == "zh"]
+                hint = f"; put the Chinese in role '{others[0]}'" if others else ""
+                errors.append(ValidationError(
+                    path, f"scene '{sid}': role '{role}' must be English only (no CJK characters){hint}",
+                ))
+            elif lang == "zh" and not _contains_cjk(content):
+                errors.append(ValidationError(
+                    path, f"scene '{sid}': role '{role}' must be Chinese (the translation of this page)",
+                ))
+    return errors
+
+
 # ════════════════════════════════════════════════════════════════════════════
 # Aggregator presets
 # ════════════════════════════════════════════════════════════════════════════
@@ -1188,6 +1259,7 @@ def validate_integrity(dsl: dict) -> list[ValidationError]:
     # 是**渲得出来但错位**,只有在真要渲之前说才来得及。
     errors += _check_step_narration_alignment(dsl)
     errors += _check_slide_cross_field(dsl)
+    errors += _check_text_layer_rules(dsl)
     return errors
 
 
@@ -1222,4 +1294,5 @@ def validate_dsl(dsl: dict) -> list[ValidationError]:
     errors += _check_line_sync_narration(dsl)
     errors += _check_step_narration_alignment(dsl)
     errors += _check_slide_cross_field(dsl)
+    errors += _check_text_layer_rules(dsl)
     return errors

package/skills/template-registry/video_dsl/runtime/timeline_compiler.py

@@ -19,13 +19,23 @@ from typing import Optional
 
 CHARS_PER_SECOND = 5.0
 
-_PRIMARY_BREAK = re.compile(r"(?<=[。！？；\n])")
+# 英文句末：`.!?`（可跟一个收尾引号/括号）且后面是空白。以前只认中文标点，英文旁白
+# 从来不在句号处断，只能落到逗号 / 30 字空格硬切，切出 `went home. A few days later,`
+# 这种跨句的行，所有段的 sentenceIndex 也全是 0。常见缩写后的句点不算句末。
+_EN_SENTENCE_END = (
+    r"(?:(?<=[.!?])|(?<=[.!?][\"'\u201d\u2019)]))"
+    r"(?<!\bMr\.)(?<!\bMs\.)(?<!\bDr\.)(?<!\bSt\.)(?<!\bMrs\.)(?<!\be\.g\.)(?<!\bi\.e\.)(?<!\bvs\.)"
+    r"(?=\s)"
+)
+_PRIMARY_BREAK = re.compile(r"(?<=[。！？；\n])|" + _EN_SENTENCE_END)
 # 次级断点**不含顿号 `、`**。顿号是并列项之间的分隔，在它上面下刀就是把一串并列
 # 拦腰切开（`调研、策划、` ／ `开发、分析，…`）。这一层只数字符、分不出标点的语义
 # 强度，所以干脆不碰它 —— 真需要再切时交给渲染端，那边是分级的（句末 > 冒号 >
 # 逗号 > 顿号 > 空格），只有前面几级都切不动才会降到顿号。
 _SECONDARY_BREAK = re.compile(r"(?<=[，,：:])")
-_PRIMARY_TAIL = re.compile(r"[。！？；\n]\s*$")
+_PRIMARY_TAIL = re.compile(
+    r"(?:[。！？；\n]|(?<!\bMr)(?<!\bMs)(?<!\bDr)(?<!\bSt)(?<!\bMrs)(?<!\bvs)[.!?][\"'\u201d\u2019)]?)\s*$"
+)
 
 # ─── 这两个上限管的是「喂给 TTS 的行」，不再是「一屏字幕」 ────────────────
 #