@remixmate/cli 0.9.72 → 0.9.74

package/skills/gen-script/skill.json

@@ -56,6 +56,16 @@
         "type": "string",
         "description": "Template id (from template_registry). The template owns outputLanguage and may also declare a defaultVoiceId; both flow into the produced DSL."
       },
+      "template_source": {
+        "type": "string",
+        "enum": ["user-explicit", "user-confirmed", "auto"],
+        "description": "How template_id was chosen; recorded at meta.templateSelection.source. **Always pass it with template_id.** `user-explicit`: the user named this template — its id or name, a pasted template detail-page prompt or link, or it was settled earlier in this conversation. `user-confirmed`: the user did not name one, you offered candidates and the user picked or approved this one in the chat. `auto`: you picked it yourself and the user has not confirmed — gen_script refuses, so instead of calling it, show the candidates and end the turn. Never mark a template user-confirmed that the user has not seen and approved."
+      },
+      "template_candidates": {
+        "type": "array",
+        "items": { "type": "string" },
+        "description": "The template ids you offered the user, recommended first; recorded at meta.templateSelection.candidates. Pass with template_source=user-confirmed — template_id must then be one of them (a template the user named outside the shortlist is user-explicit)."
+      },
       "source_document_url": {
         "type": "string",
         "description": "Source document URL (.pptx) for a template that builds its per-scene picture from that document (capabilities.payloadHydration). **Must** pass it for such a template: it is recorded at meta.sourceDocument, and prepare_video_assets then fetches the layout itself and fills the customPayload field the template declares (payloadHydration.payloadField) on every scene. Pass the same URL you gave doc_parse. You are NOT expected to author or even read that layout data — one page is hundreds of KB, so it never travels through the conversation; without this reference the DSL is rejected downstream after you have already written all the narration."
@@ -128,6 +138,8 @@
       "variant",
       "background_preset",
       "template_id",
+      "template_source",
+      "template_candidates",
       "source_document_url",
       "headline",
       "subheadline",

package/skills/gen-script/version.json

@@ -2,6 +2,6 @@
   "skillName": "gen-script",
   "repoName": "agent-skill-media-maker",
   "skillId": "474",
-  "version": "V18",
+  "version": "V19",
   "skillDescription": "Video-script generation skill. Turns a user-supplied topic into a structured Video DSL (JSON) that describes the full video — scene structure, asset requirements, and narrative flow.\n\nUse this skill as soon as the user mentions any of these intents:\n- Write a video script, generate a video script, plan a video, write storyboards\n- Create a short video, plan video content, generate a Video DSL\n- Turn a topic into a video structure / video content plan\n\nEven when the user does not say \"generate the DSL\", use this skill whenever they want to turn a topic into a structured video plan.\n\n⚠️ Stop-and-confirm gate: after this skill returns a DSL, show the full script and wait for the user's explicit confirmation. Never call `prepare_video_assets` in the same turn."
 }

package/skills/prepare-video-assets/scripts/prepare_video_assets.py

@@ -1,21 +1,21 @@
 #!/usr/bin/env python3
-"""Thin wrapper that delegates to the canonical render_video.py with --resolve-only forced on.
 
-The two-phase video pipeline shares one Python implementation:
 
-  prepare_video_assets  →  render_video.py --resolve-only ...  (this file)
-  render_video          →  render_video.py ...                 (sibling skill)
 
-By exec'ing the canonical script we get verbatim stdout / stderr / exit-code / signal forwarding
-without any subprocess wrapper layer. The agent layer's stdout parsing for the
-"📦 render job jobId: N" line therefore continues to work unchanged.
 
-The wrapper unconditionally injects --resolve-only as the first argument. The MCP tool schema for
-this skill excludes job_id / render_plan / resolve_only / upload_title / no_upload / renderer, so
-the LLM cannot push this entry point into Phase 3 territory; even if a shell caller passes those
-flags directly to this wrapper, --resolve-only still wins because asset-resolve-and-exit always
-short-circuits before the render step in render_video.py.
-"""
+
+
+
+
+
+
+
+
+
+
+
+
+
 from __future__ import annotations
 
 import os
@@ -23,14 +23,14 @@ import sys
 
 
 def _canonical_script_path() -> str:
-    """Resolve the absolute path to the render-video skill's render_video.py.
 
-    Layout (relative to this file):
-        apps/ab-skill/skills/prepare-video-assets/scripts/prepare_video_assets.py  <-- HERE
-        apps/ab-skill/skills/render-video/scripts/render_video.py                  <-- target
 
-    The relative jump is ../../render-video/scripts/render_video.py.
-    """
+
+
+
+
+
+
     here = os.path.dirname(os.path.abspath(__file__))
     return os.path.normpath(
         os.path.join(here, "..", "..", "render-video", "scripts", "render_video.py")
@@ -40,7 +40,7 @@ def _canonical_script_path() -> str:
 def main() -> None:
     script = _canonical_script_path()
     if not os.path.isfile(script):
-        # Fail fast with a human-readable message rather than letting execv raise OSError.
+
         print(
             f"❌ prepare_video_assets wrapper could not locate the canonical render script.\n"
             f"   Expected at: {script}\n"
@@ -48,7 +48,7 @@ def main() -> None:
             file=sys.stderr,
         )
         sys.exit(2)
-    # Force --resolve-only into argv before any user-supplied flag.
+
     forwarded = ["--resolve-only", *sys.argv[1:]]
     os.execv(sys.executable, [sys.executable, script, *forwarded])
 

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

@@ -1,11 +1,11 @@
-"""Chrome Headless vendor sync — separated from render_video.py.
 
-把 Chrome 下载/解压/共享缓存这块"部署/运维相关"的逻辑收拢到一个文件，让
-render_video.py 只剩"调度 + 渲染"。日后如果要把这块搬到 ab-render Dockerfile
-或独立 prepare 脚本，只需要换掉这个模块的实现，render_video 不用动。
 
-公共入口只有一个：sync_chrome_headless_vendor()。其它都是辅助函数，给 sync 用。
-"""
+
+
+
+
+
+
 from __future__ import annotations
 
 import builtins
@@ -13,7 +13,7 @@ import json
 import os
 import platform
 import shutil
-import subprocess  # noqa: F401  (kept for parity with original imports)
+import subprocess
 import sys
 import tempfile
 import urllib.error
@@ -29,14 +29,14 @@ def _now_iso() -> str:
 
 
 def _log(msg: str) -> None:
-    """Tiny stderr-with-timestamp helper, mirrors render_video.LogPrint format."""
+
     now = datetime.now()
     stamp = now.strftime("%Y%m%d %H%M%S") + f":{now.microsecond // 1000:03d}"
     builtins.print(f"[{stamp}] {msg}", file=sys.stderr)
 
 
 def _detect_remotion_chrome_platform() -> Optional[str]:
-    """与 Remotion BrowserFetcher 一致的平台标识；不支持时返回 None。"""
+
     system = platform.system()
     machine = platform.machine().lower()
     if system == "Darwin":
@@ -118,7 +118,7 @@ def _chrome_headless_share_zip_path(share_root: str, rem_platform: str) -> str:
 
 
 def _share_chrome_zip_ready(share_root: str, rem_platform: str, expected_version: str) -> bool:
-    """共享目录仅保留 VERSION + 各平台 zip 时，判断是否可用。"""
+
     version_path = os.path.join(share_root, "VERSION")
     if not os.path.isfile(version_path):
         return False
@@ -136,7 +136,7 @@ def _share_chrome_zip_ready(share_root: str, rem_platform: str, expected_version
 
 
 def _extract_chrome_headless_zip_to_dest(zip_path: str, dest_root: str, rem_platform: str) -> None:
-    """将官方 headless shell zip 解压到 Remotion 缓存根目录下的平台子目录。"""
+
     platform_dest = os.path.join(dest_root, rem_platform)
     if os.path.isdir(platform_dest):
         shutil.rmtree(platform_dest)
@@ -146,13 +146,13 @@ def _extract_chrome_headless_zip_to_dest(zip_path: str, dest_root: str, rem_plat
 
 
 def sync_chrome_headless_vendor(renderer_dir: str, render_plan: dict) -> None:
-    """
-    共享目录（默认 /tmp/agent-share/chrome-headless）只存 VERSION + chrome-headless-shell-<平台>.zip；
-    若 zip 与版本匹配则本地解压到 node_modules/.remotion/chrome-headless-shell/，否则从 OSS 下载，
-    成功后把 zip 写入共享目录（不存解压后的目录树）。
 
-    所有 progress / warning 经 stderr 打印；render_plan["logs"] 上记录关键事件。
-    """
+
+
+
+
+
+
     if os.environ.get("REMOTION_SKIP_VENDOR_CHROME", "").strip().lower() in ("1", "true", "yes"):
         render_plan["logs"].append({
             "phase": "render",

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

@@ -1,16 +1,16 @@
-"""数字人连续段：分段、驱动音频规划、起播帧注入。
 
-一个 ``source=gen-digital-human`` 的 avatar 资产被**相邻**几个场景的
-``props.avatarAssetId`` 同时引用 = 一个连续段：生成一段视频，人物跨场景不重启。
-要让第 k 个场景的第 0 帧正好对上视频里的对应位置，驱动音频必须是「段内各场景
-旁白、每段补静音到场景最终时长」拼成的一条——所以数字人只能在**排轴之后**生成
-（场景时长由真实 TTS 时长决定，见 render_video.adjust_timeline_to_audio）。
 
-本模块只放纯函数与 ffmpeg 拼接；调度（何时生成、如何回写 RenderPlan 状态）在
-render_video.resolve_digital_human_segments。全部按 RenderPlan 的通用字段工作，
-不识别具体模板：模板是否支持由 template.json 的 ``capabilities.digitalHuman`` 声明，
-写没写 avatar 资产由上游决定。
-"""
+
+
+
+
+
+
+
+
+
+
+
 
 from __future__ import annotations
 
@@ -21,10 +21,10 @@ from typing import Optional
 
 DIGITAL_HUMAN_SOURCE = "gen-digital-human"
 
-#: 驱动音频下限。飞影要求 5 秒 ～ 30 分钟；不足时在段尾补静音。
+
 MIN_DRIVE_SEC = 5.0
 
-#: 拼接输出的采样率 / 声道。口型驱动不需要立体声，统一成单声道避免混格式失败。
+
 _SAMPLE_RATE = 44100
 
 
@@ -42,11 +42,11 @@ def digital_human_asset_ids(render_plan: dict) -> set[str]:
 
 
 def collect_segments(render_plan: dict) -> dict[str, list[int]]:
-    """avatar 资产 id → 引用它的 timeline 下标（按出场顺序）。
 
-    只收 gen-digital-human 资产：``existing`` 的数字人视频是作者自备的，起播点由作者
-    自己负责（``digitalHuman.startFromSec``），管线不去猜。
-    """
+
+
+
+
     dh_ids = digital_human_asset_ids(render_plan)
     segments: dict[str, list[int]] = {}
     for idx, entry in enumerate(render_plan.get("timeline", [])):
@@ -69,11 +69,11 @@ def segment_role(pos: int, count: int) -> str:
 
 
 def inject_segment_props(render_plan: dict, segments: dict[str, list[int]]) -> list[str]:
-    """给段内每个场景写 ``avatarOffsetFrames`` / ``avatarSegmentRole``。
 
-    偏移用 RenderPlan 里最终的 ``durationFrames`` 累加，不按秒重算，避免累计舍入。
-    返回告警（不相邻的同一资产：偏移仍按出场顺序累加，但中间隔开的场景会让口型错位）。
-    """
+
+
+
+
     warnings: list[str] = []
     timeline = render_plan.get("timeline", [])
     for asset_id, indices in segments.items():
@@ -94,11 +94,11 @@ def inject_segment_props(render_plan: dict, segments: dict[str, list[int]]) -> l
 
 
 def plan_drive_pieces(render_plan: dict, indices: list[int]) -> list[dict]:
-    """段内每个场景一块：``{"sceneId", "url" | None, "durationSec"}``。
 
-    ``url`` 取场景旁白资产（``props.narrationAssetId``）已生成的地址；没有旁白的场景
-    是整段静音。块长 = 场景最终时长；总长不足 MIN_DRIVE_SEC 时把最后一块拉长。
-    """
+
+
+
+
     fps = (render_plan.get("renderConfig") or {}).get("fps") or 30
     assets = {a.get("assetId"): a for a in render_plan.get("resolvedAssets", [])}
     timeline = render_plan.get("timeline", [])
@@ -119,10 +119,10 @@ def plan_drive_pieces(render_plan: dict, indices: list[int]) -> list[dict]:
 
 
 def direct_drive_url(render_plan: dict, indices: list[int]) -> Optional[str]:
-    """单场景段、且旁白本身就够长时，直接拿旁白音频驱动，不必拼接。
 
-    视频比场景短几百毫秒（旁白尾巴）无妨：模板默认 ``onEnd: freeze`` 停在最后一帧。
-    """
+
+
+
     if len(indices) != 1:
         return None
     entry = render_plan.get("timeline", [])[indices[0]]
@@ -137,11 +137,11 @@ def direct_drive_url(render_plan: dict, indices: list[int]) -> Optional[str]:
 
 
 def split_segment_per_scene(render_plan: dict, asset_id: str, indices: list[int]) -> list[str]:
-    """把一个连续段拆成每场景一段（拿不到 ffmpeg 拼接时的降级）。
 
-    人物会在场景边界重新起播，但每段都能直接用本场景旁白驱动，口型不会错位。
-    新资产 id = ``<原 id>.<sceneId>``；原资产从 resolvedAssets 移除，场景改指新 id。
-    """
+
+
+
+
     assets = render_plan.get("resolvedAssets", [])
     original = next((a for a in assets if a.get("assetId") == asset_id), None)
     if original is None:
@@ -160,7 +160,7 @@ def split_segment_per_scene(render_plan: dict, asset_id: str, indices: list[int]
 
 
 def build_concat_command(pieces: list[dict], out_path: str, ffmpeg: str = "ffmpeg") -> list[str]:
-    """ffmpeg 参数：每块重采样成单声道 → 补静音 → 精确截到块长 → 顺序拼接成 mp3。"""
+
     cmd = [ffmpeg, "-y", "-loglevel", "error"]
     filters: list[str] = []
     labels: list[str] = []
@@ -190,7 +190,7 @@ def find_ffmpeg() -> Optional[str]:
 
 
 def concat_drive_audio(pieces: list[dict], out_dir: str, name: str, ffmpeg: str, timeout: int = 300) -> str:
-    """拼出驱动音频，返回本地文件路径。失败抛 RuntimeError。"""
+
     os.makedirs(out_dir, exist_ok=True)
     safe = "".join(ch if ch.isalnum() or ch in "-_." else "_" for ch in name)
     out_path = os.path.join(out_dir, f"dh-drive-{safe}.mp3")

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

@@ -1,15 +1,15 @@
-"""Video duration probe — separated from render_video.py.
 
-screen-walkthrough 这种"按 TTS 时间轴 + 视频素材池"的模板需要知道远端视频的
-真实时长才能选出正确的 adaptStrategy（fit-exact / trim-and-speed / freeze-tail
-/ static-fallback）。这里给出三条策略，按速度排序：
 
-  1. 通过 mediaId（从 CDN URL 中切出）查 ab-api /file/get
-  2. HEAD 请求按 Content-Length 估算（粗略）
-  3. 下载前 2 MB 用 ffprobe 解析（精确但贵）
 
-公共入口只有一个：probe_video_duration(url) -> Optional[float]。
-"""
+
+
+
+
+
+
+
+
+
 from __future__ import annotations
 
 import builtins
@@ -31,12 +31,12 @@ def _log(msg: str) -> None:
 
 
 def _extract_media_id_from_url(url: str) -> Optional[str]:
-    """Extract mediaId (vodVideoId) from CDN URL.
 
-    Supported formats:
-      - http://cdn.aibearer.com/{mediaId}/{hash}-ld.mp4
-      - https://cdn.aibearer.com/{mediaId}/...
-    """
+
+
+
+
+
     try:
         from urllib.parse import urlparse
         parsed = urlparse(url)
@@ -51,10 +51,10 @@ def _extract_media_id_from_url(url: str) -> Optional[str]:
 
 
 def _query_duration_from_api(url: str) -> Optional[float]:
-    """Query ab-api /file/get by mediaId to get the precise duration from database.
 
-    Returns duration in seconds, or None if query fails or duration not available.
-    """
+
+
+
     media_id = _extract_media_id_from_url(url)
     if not media_id:
         return None
@@ -99,15 +99,15 @@ def _query_duration_from_api(url: str) -> Optional[float]:
 
 
 def probe_video_duration(url: str) -> Optional[float]:
-    """Probe the duration (in seconds) of a remote video URL.
 
-    Strategy (ordered by priority):
-      1. Query ab-api /file/get by mediaId extracted from CDN URL (fastest, most accurate).
-      2. HEAD request to estimate from Content-Length.
-      3. Download first 2MB + ffprobe (if available).
 
-    Returns duration in seconds, or None on failure.
-    """
+
+
+
+
+
+
+
     api_duration = _query_duration_from_api(url)
     if api_duration is not None:
         return api_duration
@@ -178,16 +178,16 @@ def probe_video_duration(url: str) -> Optional[float]:
 
 
 def probe_video_meta(url: str, timeout: float = 20.0) -> dict:
-    """Probe a remote video's pixel size and duration with ffprobe (reads the URL directly).
 
-    Returns ``{"width": int, "height": int, "duration_ms": int}`` with only the fields it
-    could read; ``{}`` when ffprobe is missing or the probe fails. Callers treat every
-    field as optional.
 
-    为什么要：数字人盒子的宽高比缺省按 3:4 算，而克隆素材多是 9:16 左右的竖版
-    （实测 720×1264）。资产上没有宽高时，模板用 3:4 的盒子 object-fit: cover 裁掉
-    人物的头顶和下半身（conv_XKTQxEBLdZy3i_jIB7IG）。
-    """
+
+
+
+
+
+
+
+
     try:
         result = subprocess.run(
             [
@@ -212,7 +212,7 @@ def probe_video_meta(url: str, timeout: float = 20.0) -> dict:
         s = streams[0]
         w, h = s.get("width"), s.get("height")
         if isinstance(w, int) and isinstance(h, int) and w > 0 and h > 0:
-            # 手机素材常带 90° / 270° 旋转元数据：播放器按旋转后的方向显示。
+
             rotation = 0
             for side in s.get("side_data_list") or []:
                 if isinstance(side.get("rotation"), (int, float)):

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

@@ -1,12 +1,12 @@
-"""VOD playback URL polling — separated from render_video.py.
 
-ab-render 通过远程渲染服务把视频上传到阿里云 VOD 后，会立即返回一个
-`vod://` 协议地址 + fileId。但前端需要的是 CDN 加速地址（`http://cdn.aibearer.com/...`），
-这要等阿里云转码回调到达 ab-api、回写 `file.fileUrl` 之后才能拿到。
 
-这个模块负责轮询 ab-api /file/get 直到拿到 CDN URL（或超时返回 OSS 原始地址）。
-公共入口只有一个：poll_vod_playback_url(file_id, priv_token, ...) -> Optional[str]。
-"""
+
+
+
+
+
+
+
 from __future__ import annotations
 
 import builtins
@@ -33,13 +33,13 @@ def poll_vod_playback_url(
     max_wait: float = 60.0,
     interval: float = 3.0,
 ) -> Optional[str]:
-    """轮询 ab-api /file/get，等待 VOD CDN URL 就绪。
 
-    VOD 上传后通常 10-30 秒完成转码。此函数策略：
-      - 一旦看到非 `vod://` 且非阿里云 OSS 域名的 URL（说明 CDN 转码完成），立即返回
-      - 看到阿里云 OSS 直链时记为 fallback，继续等 CDN
-      - 超时仍未拿到 CDN 时退回到 OSS 原始地址
-    """
+
+
+
+
+
+
     api_base = os.environ.get(
         "MM_API_BASE_URL",
         os.environ.get("MM_BACKEND_API_URL", "https://api.remixmate.ai/api"),

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

@@ -1,28 +1,28 @@
 #!/usr/bin/env python3
-"""
-远程渲染 HTTP 客户端 — 对接独立部署的 remotion-renderer 服务。
-
-接口（位于 /remotion-renderer/server/index.ts）：
-  - POST /render        → { code, data: { taskId } }
-        body 可选 coverCompositionId（如 "SpotlightCardCover"）；传了的话视频
-        渲染完成后服务端会再串行渲一帧静态封面、一并上传。
-  - POST /renderStatus  → { code, data: { status, progress, fileUrl?,
-                                          coverUrl?, coverError?, error? } }
-        coverUrl  视频成功且封面成功时存在
-        coverError 视频成功但封面失败时存在（视频本身仍然 succeeded）
-
-认证头（复用 X-Priv-Token 规范，服务端通过后端 API 校验）：
-  X-Priv-Token: <PRIV_TOKEN>
-  x-invoke-skill:         render-video
-  x-invoke-agent:         <AGENT_NAME>  # optional
-
-Base URL 优先级：
-  1. 调用方显式传入 base_url
-  2. 环境变量 REMOTION_RENDER_API_URL（指向本地/staging 时在此覆盖）
-  3. 默认 https://api-render.remixmate.ai（生产，零配置可用）
-
-本模块只负责 HTTP 层；调用方（render_video.py）负责把结果写回 render_plan。
-"""
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
 
 from __future__ import annotations
 
@@ -61,16 +61,16 @@ TERMINAL_FAILURE = {STATUS_FAILED, STATUS_CANCELLED}
 
 
 class RemoteRenderError(RuntimeError):
-    """抛出至调用方，由上层写入 render_plan.errors。"""
+    pass
 
 
 class RemoteRenderTimeout(RemoteRenderError):
-    """轮询超时 —— 任务**没有失败**，只是我们不等了。
 
-    与普通 RemoteRenderError 分开，是因为上层的处置完全相反：
-    真失败 → 可以重新提交一个任务；超时 → 后端还在渲，重新提交等于白烧一遍，
-    必须保留 taskId 让下一次调用续上。
-    """
+
+
+
+
+
 
     def __init__(self, message: str, task_id: str):
         super().__init__(message)
@@ -144,7 +144,7 @@ def start_render(
     conversation_id: Optional[str] = None,
     path: str = "/render",
 ) -> str:
-    """POST <path>（默认 /render，私有模板动态渲染传 /renderDraft）；返回 taskId。"""
+
     result = _request(
         "POST",
         path,
@@ -174,16 +174,16 @@ def poll_render(
     adaptive_interval: bool = False,
     status_path: str = "/renderStatus",
 ) -> dict:
-    """POST <status_path>（默认 /renderStatus，私有模板传 /renderDraftStatus）轮询直到完成/失败/超时。
 
-    成功时返回 data dict（至少包含 fileUrl），否则抛 RemoteRenderError。
-    on_progress 回调在每次成功请求后触发，参数为完整 data dict。
 
-    adaptive_interval=True 时根据渲染进度自动调整轮询间隔：
-      0-20%: max(interval * 0.5, 1.0)（起步阶段采样加密，短渲染也能看到进度）
-      20-80%: interval（正常渲染阶段）
-      80-100%: max(interval * 0.6, 1.0)（编码收尾阶段，更频繁检查）
-    """
+
+
+
+
+
+
+
+
     start = time.monotonic()
     consecutive_errors = 0
     current_progress = 0.0
@@ -233,7 +233,7 @@ def poll_render(
         if on_progress is not None:
             try:
                 on_progress(data)
-            except Exception:  # noqa: BLE001 — 进度回调异常不应中断轮询
+            except Exception:
                 pass
 
         status = str(data.get("status") or "").lower()
@@ -243,11 +243,11 @@ def poll_render(
             err = data.get("error") or data.get("errorMsg") or data.get("message") or "unknown"
             raise RemoteRenderError(f"remote render failed ({status}): {err}")
 
-        # Adaptive interval: adjust sleep based on current progress
+
         current_progress = float(data.get("progress") or 0.0)
         if adaptive_interval:
             if current_progress < 0.2:
-                # 起步阶段(含服务端很快就渲完的短片)采样加密，避免 0%→完成的跳变
+
                 sleep_time = max(interval * 0.5, 1.0)
             elif current_progress >= 0.8:
                 sleep_time = max(interval * 0.6, 1.0)