media-router

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Media Router

媒体路由

Follow shared release-shell rules in:
  • postplus-shared
    release-shell rules
Use this skill when the user wants to do something with:
  • audio
  • video
  • transcripts
  • subtitles
  • A-roll / B-roll edit prep
This skill is a routing layer.
It should decide:
  • what the actual task is
  • what output shape is required
  • whether the job is rough, subtitle-ready, or edit-ready
  • whether cost or quality should dominate
  • which downstream skill or skill chain should execute
This skill should not pretend every request is "just transcription".
请遵循以下共享发布规则:
  • postplus-shared
    发布规则
当用户需要处理以下内容时,可使用此Skill:
  • 音频
  • 视频
  • 转录文本
  • 字幕
  • A-roll / B-roll 编辑准备
此Skill是一个路由层。
它需要决定:
  • 实际任务是什么
  • 需要的输出形式
  • 任务是粗略处理、字幕就绪还是可编辑就绪
  • 应优先考虑成本还是质量
  • 应执行哪个下游Skill或Skill链
此Skill不应将所有请求都当作“只是转录”来处理。

Use For

使用场景

  • deciding whether to call audio STT or video STT
  • deciding when timestamps are required
  • deciding when subtitle packaging is enough
  • deciding when semantic video understanding must be added
  • deciding when editing-decision-engine should run
  • deciding whether a batch workflow should be used
  • 决定调用音频STT还是视频STT
  • 决定是否需要时间戳
  • 决定是否仅需字幕打包即可
  • 决定是否必须添加语义视频理解
  • 决定是否应运行编辑决策引擎
  • 决定是否应使用批处理工作流

Read These References

参考文档

  • references/brief-schema.md
  • references/routing-modes.md
  • references/brief-schema.md
  • references/routing-modes.md

Core Rule

核心规则

Route by job intent, not by input file type alone.
A video input may still be:
  • plain transcription
  • subtitle generation
  • transcript plus semantic understanding
  • edit-beat planning
The route should be chosen from five dimensions:
  • input type
  • output goal
  • required precision
  • scale
  • cost mode
根据任务意图进行路由,而非仅根据输入文件类型。
视频输入可能对应以下任务:
  • 纯转录
  • 字幕生成
  • 转录文本加语义理解
  • 编辑节拍规划
路由需从五个维度选择:
  • 输入类型
  • 输出目标
  • 所需精度
  • 规模
  • 成本模式

Default Routing Logic

默认路由逻辑

audio -> transcript

audio -> transcript

Use
skills/40-creative/audio-transcription
.
使用
skills/40-creative/audio-transcription

video -> transcript/subtitles

video -> transcript/subtitles

Use
skills/40-creative/video-transcription
.
使用
skills/40-creative/video-transcription

transcript with timestamps -> subtitle files

transcript with timestamps -> subtitle files

Use
skills/40-creative/subtitle-packager
.
使用
skills/40-creative/subtitle-packager

transcript + assets -> cut logic / B-roll plan

transcript + assets -> cut logic / B-roll plan

Use
skills/40-creative/editing-decision-engine
.
使用
skills/40-creative/editing-decision-engine

video -> edit-ready plan

video -> edit-ready plan

Chain:
  1. video-transcription
  2. subtitle-packager
    if subtitle files are needed
  3. frame-extraction
    if visual-proof snapshots are needed
  4. editing-decision-engine
链式流程:
  1. video-transcription
  2. 若需要字幕文件则调用
    subtitle-packager
  3. 若需要视觉验证快照则调用
    frame-extraction
  4. editing-decision-engine

Quality Thresholds

质量阈值

rough

rough
(粗略处理)

Good enough for:
  • search
  • quick review
  • rough logging
  • candidate beat spotting
Prefer:
  • cheaper models
  • fewer enrichments
适用于:
  • 搜索
  • 快速审核
  • 粗略日志记录
  • 候选节拍识别
优先选择:
  • 低成本模型
  • 更少的增强处理

subtitle-ready

subtitle-ready
(字幕就绪)

Good enough for:
  • human-readable captions
  • SRT / VTT generation
  • reliable sentence or word timestamps
Prefer:
  • Whisper with timestamps
适用于:
  • 人类可读字幕
  • SRT / VTT 生成
  • 可靠的句子或单词级时间戳
优先选择:
  • 带时间戳的Whisper模型

edit-ready

edit-ready
(可编辑就绪)

Good enough for:
  • A-roll/B-roll decision making
  • cutaway placement
  • beat-level planning
Prefer:
  • Whisper with timestamps
  • semantic video understanding when proof footage matters
  • editing-decision-engine downstream
适用于:
  • A-roll/B-roll决策
  • 切换镜头放置
  • 节拍级规划
优先选择:
  • 带时间戳的Whisper模型
  • 当需要验证素材时使用语义视频理解
  • 下游调用编辑决策引擎

User-Facing Explanation Rule

用户端说明规则

Do not expose internal route labels unless the user is explicitly designing the system.
Normal user-facing language should say:
  • what will be extracted first
  • what will be aligned second
  • what editing decisions become possible afterward
除非用户明确在设计系统,否则不要暴露内部路由标签。
面向普通用户的语言应说明:
  • 首先将提取什么内容
  • 接下来将对齐什么内容
  • 之后可实现哪些编辑决策

Default Output

默认输出

Return:
  • chosen route
  • why it fits the task
  • primary skill
  • supporting skills
  • model plan
  • expected output artifacts
  • whether timestamps are mandatory
返回:
  • 选定的路由
  • 该路由适配任务的原因
  • 主Skill
  • 辅助Skill
  • 模型方案
  • 预期输出产物
  • 是否必须包含时间戳