@remixmate/cli 0.9.45 → 0.9.46

package/dist/manifest.json

@@ -1,7 +1,7 @@
 {
   "schemaVersion": 1,
-  "version": "0.9.45",
-  "generatedAt": "2026-09-23T05:34:40.922Z",
+  "version": "0.9.46",
+  "generatedAt": "2026-09-23T08:47:39.701Z",
   "skills": [
     {
       "id": "doc-parse",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@remixmate/cli",
-  "version": "0.9.45",
+  "version": "0.9.46",
   "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",

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

@@ -328,7 +328,11 @@ def plan_scenes(
     opening_duration = max(3, round(duration * 0.12))
     ending_duration = max(3, round(duration * 0.12))
     content_total = duration - opening_duration - ending_duration
-    per_content = max(3, round(content_total / content_scenes))
+    # 余数摊到前几段，保证各段加总正好等于 duration。以前是 round(均值) × 段数：
+    # 2026-09-23 线上 ppt-to-video 传 --duration 600 --scenes 23，21 段 × round(21.7)=22
+    # 加总 606s，撞上模板 supportedDurations.max=600 被自己的时长门禁拒掉。
+    # 每段仍有 3s 下限 —— 段数多到摊不开时加总会超出，那是真超出，交给门禁报。
+    base_content, extra_content = divmod(max(content_total, 0), content_scenes)
 
     scenes.append({
         "purpose": "opening",
@@ -339,7 +343,7 @@ def plan_scenes(
     for i in range(content_scenes):
         scenes.append({
             "purpose": "point",
-            "duration": per_content,
+            "duration": max(3, base_content + (1 if i < extra_content else 0)),
             "label": _localize_label(language, "label_point", n=i + 1),
         })
 
@@ -617,6 +621,7 @@ def _resolve_contract(template_config: dict | None) -> dict:
         "payload_defaults": payload_defaults,
         "primary_visual_type": primary_visual_type,
         "default_layout": caps.get("defaultLayout"),
+        "authored_scenes": caps.get("authoredScenes") is True,
     }
 
 
@@ -791,6 +796,16 @@ def _plan_contract_scenes(
 
     fixed = contract["fixed_scenes"] or [{"purpose": "opening"}]
     n = len(fixed)
+    # 非 arc 策略的场景数由模板钉死，--scenes 不生效 —— 但必须出声。2026-09-23 线上
+    # screen-walkthrough 传 --scenes 6 拿回 2 场，没有任何提示，模型只能自己猜为什么。
+    if scene_count is not None and scene_count != n:
+        print(
+            f"⚠️  requested {scene_count} scene(s) but this template's sceneStrategy="
+            f"{contract['scene_strategy']} always plans {n}: --scenes is ignored here.\n"
+            f"   The DSL below has {n} scenes — read the count back from it before showing "
+            f"the script to the user.",
+            file=sys.stderr,
+        )
     per = max(1, round(duration / n))
     plans = []
     for i, fs in enumerate(fixed):
@@ -1709,6 +1724,20 @@ def build_dsl(
 
     _record_source_document(dsl, template_config, template_id, source_document_url)
 
+    if contract["authored_scenes"]:
+        # 场景表由调用方按用户素材编写（如按操作时间点切录屏），骨架只提供 meta/global。
+        # 这个标记必须由骨架携带：ab-agent 看到它才会整份采用模型提交的 DSL，
+        # 否则它以骨架为基底合并，新增场景、existing 素材、visuals 都会被静默丢掉。
+        dsl["meta"]["authoredScenes"] = True
+        print(
+            f"ℹ️  template '{template_id}' declares capabilities.authoredScenes: the scene list below "
+            f"is only a placeholder.\n"
+            f"   Write the full DSL yourself from the user's material (scenes, assets, visuals, "
+            f"customPayload) and submit all of it to prepare_video_assets — it is used as-is, "
+            f"not merged onto this skeleton. Keep meta.templateId.",
+            file=sys.stderr,
+        )
+
     _enforce_supported_duration(dsl, template_config, template_id)
 
     return dsl

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

@@ -110,7 +110,10 @@ def _rebalance_subtitle_segments(segs: list[dict]) -> list[dict]:
         for p in parts:
             ratio = len(p) / total_chars
             frames = max(int(total_frames * ratio), 1)
+            # `**seg` 而不是只写三个键：切出来的各部分仍属于**同一句**，
+            # 所以 sentenceIndex 等字段要原样继承（见 build_render_plan 处的说明）。
             expanded.append({
+                **seg,
                 "text": _strip_subtitle_trailing_punct(p.strip()),
                 "startFrame": cur_frame,
                 "endFrame": cur_frame + frames,
@@ -848,9 +851,13 @@ def build_render_plan(dsl: dict, binding: dict) -> dict:
             local_subs = split_subtitle_from_lines(lines, duration_frames, fps)
         else:
             local_subs = split_subtitle(narration_text, duration_frames, fps)
+        # ⚠️ 平移帧号时**必须保留原有字段**（`**s`），不要只挑三个键重建。
+        # 这里曾经只搬 text/startFrame/endFrame，于是 timeline_compiler 给每段打上的
+        # `sentenceIndex`（模板用它把动作对齐到**句**而不是段）在这一步被静默丢掉 ——
+        # 下游模板拿不到句边界，只能退回按段对齐，也就是那个错位 bug 本身。
         subtitle_segments = [
             {
-                "text": s["text"],
+                **s,
                 "startFrame": s["startFrame"] + current_frame,
                 "endFrame": s["endFrame"] + current_frame,
             }
@@ -1519,6 +1526,7 @@ def adjust_timeline_to_audio(render_plan: dict) -> dict:
                             # 起点为终点。
                             end_f = scene_dur if i == len(old_subs) - 1 else starts[i + 1]
                             new_subs.append({
+                                **s,
                                 "text": text,
                                 "startFrame": current_frame + start_f,
                                 "endFrame": current_frame + end_f,
@@ -1538,6 +1546,7 @@ def adjust_timeline_to_audio(render_plan: dict) -> dict:
                             ratio = _plain(text) / total_chars
                             seg_frames = max(int(sub_span * ratio), fps // 2)
                             new_subs.append({
+                                **s,
                                 "text": text,
                                 "startFrame": current_frame + cursor,
                                 "endFrame": current_frame + cursor + seg_frames,
@@ -1560,7 +1569,7 @@ def adjust_timeline_to_audio(render_plan: dict) -> dict:
                         rel_start = s["startFrame"] - old_scene_start
                         rel_end = s["endFrame"] - old_scene_start
                         new_subs.append({
-                            "text": s["text"],
+                            **s,
                             "startFrame": current_frame + rel_start,
                             "endFrame": current_frame + rel_end,
                         })

package/skills/render-video/version.json

@@ -2,6 +2,6 @@
   "skillName": "render-video",
   "repoName": "agent-skill-media-maker",
   "skillId": "473",
-  "version": "V27",
+  "version": "V28",
   "skillDescription": "Final-render skill: loads a persisted RenderPlan by `job_id` and drives the Remotion engine to produce the final video.\n\nUse this skill as soon as the user mentions any of these intents (after assets are already prepared):\n- Render the video, composite the video, export the video\n- Turn the prepared assets into the final clip\n- Render with Remotion\n\nPrerequisite: assets must already be generated via `prepare_video_assets`. This skill never resolves or regenerates assets — pass it a `job_id` from a previous `prepare_video_assets` call.\n\n⚠️ Stop-and-confirm gate: never call this skill until the user has explicitly confirmed the assets prepared by `prepare_video_assets`. If those assets were prepared in the current turn and the user has not replied since, stop and ask instead of rendering."
 }

package/skills/template-registry/scripts/payload_hydrator.py

@@ -56,6 +56,44 @@ SUPPORTED_SOURCES = ("doc-extract",)
 DEFAULT_INDEX_FIELD = "slideIndex"
 
 
+def declared_subfields(template_config: Optional[dict], payload_field: str) -> Optional[set[str]]:
+    """模板为 ``customPayload.<payload_field>`` 声明的子字段名集合。
+
+    注入必须按这个集合**投影**，不能把抽取器返回的整个对象原样塞进去 ——
+    两边的字段集根本不是一回事：
+
+      抽取器返回(每页)  slideIndex / title / text / rawNotes / verdict / coverage /
+                        flags / nodes / beats / mode / modeReason / ambientIds /
+                        background / recommendedSec / canvas …
+      模板声明(slide)    nodes / beats / mode / modeReason / ambientIds /
+                        background / recommendedSec / canvas
+
+    前者多出来的那些是**内容层与抽取器自己的记账字段**（备注原文、质检结论、覆盖率），
+    渲染端一个都不读。而 ``dsl_validator._check_custom_payload_fields`` 会逐字段核对
+    声明表，遇到没声明的就报「the slide does not read 'x' — the value is silently
+    dropped」并整单拒绝。2026-09-23 线上 conv_StrnauS7TaZNUQV4BfLX 就是这样：
+    注入明明成功（23/23 页），却因为多了 5 个字段被校验整单拒掉，而且**重试必然复现**。
+
+    为什么用投影而不是写个黑名单：抽取器以后再加字段（verdict / coverage 就是后加的）
+    黑名单就会再漏一次。声明表是渲染端真正读的那一份，按它投影才不会二次失效。
+
+    声明表读不到时返回 None = 不投影。这与校验层的行为一致：没有
+    ``customPayloadSchema`` 时它也不查字段名，所以不投影不会引入新的失败。
+    """
+    if not template_config:
+        return None
+    cps = template_config.get("customPayloadSchema")
+    if not isinstance(cps, dict):
+        return None
+    node = cps.get(payload_field)
+    if not isinstance(node, dict):
+        return None
+    props = node.get("properties")
+    if not isinstance(props, dict) or not props:
+        return None
+    return set(props.keys())
+
+
 class HydrationError(RuntimeError):
     """注入失败。message 必须直接告诉调用方下一步做什么。
 
@@ -206,6 +244,18 @@ def hydrate_dsl(
     filled = 0
     missing: list[str] = []
 
+    # 按模板声明投影。抽取结果里混着内容层与抽取器记账字段，渲染端一个都不读，
+    # 而校验层见到未声明字段会整单拒绝（见 declared_subfields 的说明）。
+    allowed = declared_subfields(template_config, field)
+    dropped: set[str] = set()
+
+    # 页号只有在模板**声明**了它的时候才写回去。没声明就是未知字段，写了会被校验
+    # 拒掉 —— 这正是上一版的第二个错：既投影不足，又多写了一个 customPayload.slideIndex。
+    index_is_declared = index_field in (declared_subfields(template_config, index_field) or ())
+    if not index_is_declared:
+        cps = (template_config or {}).get("customPayloadSchema")
+        index_is_declared = isinstance(cps, dict) and index_field in cps
+
     for position, scene in enumerate(scenes, start=1):
         payload = _scene_payload(scene)
         if payload.get(field) is not None:
@@ -216,9 +266,14 @@ def hydrate_dsl(
         if page is None:
             missing.append(f"scene {scene.get('id', position)} → page {page_no}")
             continue
-        payload[field] = page
-        # 把页号写回去，让 RenderPlan 里留下"这一场用的是第几页"的痕迹。
-        payload.setdefault(index_field, page_no)
+        if allowed is None:
+            payload[field] = page
+        else:
+            dropped |= {k for k in page if k not in allowed}
+            payload[field] = {k: v for k, v in page.items() if k in allowed}
+        # 留下"这一场用的是第几页"的痕迹 —— 仅当模板声明了这个字段。
+        if index_is_declared:
+            payload.setdefault(index_field, page_no)
         filled += 1
 
     source_meta = result.get("source") or {}
@@ -227,6 +282,13 @@ def hydrate_dsl(
         f"{source_meta.get('slideCount', len(pages))} page(s) of "
         f"{source_meta.get('name', 'the source document')}"
     )
+    if dropped:
+        # 说清丢了什么：这些是内容层/记账字段，渲染端本来就不读。写出来是为了
+        # 抽取器哪天把一个**该读**的字段也归到这里时，有人看得见。
+        report.append(
+            f"   （已按模板声明投影，忽略 {len(dropped)} 个渲染端不读的字段："
+            f"{', '.join(sorted(dropped)[:8])}）"
+        )
 
     if missing:
         # 页号对不上不是警告而是错误：少一页 = 成片里一个空白场景，而它渲得出来。

package/skills/template-registry/version.json

@@ -2,6 +2,6 @@
   "skillName": "template-registry",
   "repoName": "agent-skill-media-maker",
   "skillId": "475",
-  "version": "V20",
+  "version": "V21",
   "skillDescription": "Video-template registry skill. Stores every video-template definition and lists the available templates (templateId / name / aspect ratio / style tags).\n\nUse this skill as soon as the user mentions any of these intents:\n- View available templates / list every template\n\nNote: DSL→TemplateBinding is no longer a separate exposed step — once prepare_video_assets receives a template_id it builds the binding internally."
 }