broll-match-engine

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

B-roll Match Engine

B-roll匹配引擎

Follow shared release-shell rules in:

```
postplus-shared
```
release-shell rules

Use this skill when the user wants to map a talking-head edit to candidate B-roll coverage.

This skill should produce a structured match plan, not a vague creative summary.

Default output:

```
broll-plan.json
```

Optional supporting outputs:

markdown match summary
beat review table
candidate ranking table

遵循以下共享发布壳规则：

```
postplus-shared
```
发布壳规则

当用户希望将访谈式剪辑（talking-head edit）映射到候选B-roll素材时，可使用此技能。

此技能应生成结构化的匹配方案，而非模糊的创意总结。

默认输出：

```
broll-plan.json
```

可选辅助输出：

markdown匹配总结
节拍审核表格
候选素材排名表

Use For

适用场景

deciding which spoken beats need B-roll
matching a beat to the best available proof clip
deciding when to stay on A-roll instead of cutting away
identifying missing proof moments
attaching keyword-emphasis or motion-hint metadata to the match result

确定哪些口播节拍需要搭配B-roll
为节拍匹配最佳可用的证明片段
决定何时保留A-roll而非切换画面
识别缺失的证明场景
为匹配结果添加关键词强调或动态提示元数据

Trigger Signals

触发信号

Use this skill when the user asks for things like:

这些口播句子该配哪些 B-roll
帮我匹配素材
哪些地方应该切屏幕或证明画面
这条 talking head 怎么上 B-roll 更合理
给我一个 B-roll plan

Do not use this skill when the user still lacks a usable B-roll inventory.

Route that first to:

skills/40-creative/b-roll-family/broll-catalog-builder

当用户提出以下需求时使用此技能：

这些口播句子该配哪些 B-roll
帮我匹配素材
哪些地方应该切屏幕或证明画面
这条 talking head 怎么上 B-roll 更合理
给我一个 B-roll plan

当用户尚未拥有可用的B-roll素材库时，请勿使用此技能。

应先引导至：

skills/40-creative/b-roll-family/broll-catalog-builder

Read These References

参考文档

family architecture:
```
../references/skill-architecture.md
```
shared contracts:
```
../references/contracts.md
```

Use these existing local skills when helpful:

a dedicated visual analysis workflow

skills/40-creative/editing-decision-engine

家族架构：
```
../references/skill-architecture.md
```
共享协议：
```
../references/contracts.md
```

必要时可使用以下现有本地技能：

专用视觉分析工作流

skills/40-creative/editing-decision-engine

Core Principle

核心原则

Do not match by keyword overlap only.

A good match depends on:

beat meaning
proof need
visual support type
pacing need
B-roll strength

The best asset is not always the most literally similar one.

Sometimes the right choice is:

a UI proof clip
a workflow bridge clip
a pace reset clip
or no B-roll at all

请勿仅通过关键词重叠进行匹配。

优质匹配取决于：

节拍含义
证明需求
视觉支持类型
节奏需求
B-roll素材质量

最佳素材并非总是字面最相似的那一个。

有时合适的选择是：

UI证明片段
工作流程过渡片段
节奏重置片段
或完全不使用B-roll

Preferred Inputs

首选输入

Required:

```
normalized-transcript.json
```
```
broll-catalog.json
```

Strongly preferred:

```
chunked-<mode>.json
```

Optional but high-value:

A-roll semantic analysis
reference edit pattern
target platform
intended output length

必填：

```
normalized-transcript.json
```
```
broll-catalog.json
```

强烈推荐：

```
chunked-<mode>.json
```

可选但高价值：

A-roll语义分析
参考剪辑模式
目标平台
预期输出时长

Workflow

工作流程

1. Build beat units

1. 构建节拍单元

Prefer subtitle chunks or edit beats over raw transcript segments.

For each beat, capture:

spoken text
timing
beat role
visual need

优先使用字幕分段或剪辑节拍，而非原始文稿片段。

为每个节拍记录：

口播文本
时间点
节拍角色
视觉需求

2. Evaluate B-roll need

2. 评估B-roll需求

Not every beat deserves a cutaway.

For each beat, decide:

stay on face
use supporting B-roll
use primary proof B-roll
use light visual reset only

并非每个节拍都需要切换画面。

为每个节拍决定：

保留人物画面
使用辅助B-roll
使用主要证明B-roll
仅使用轻量视觉重置

3. Rank candidate assets

3. 候选素材排名

For each beat that needs B-roll, rank candidates based on:

semantic fit
support-role fit
proof strength
pacing fit
visual clarity
usable range quality

Every candidate should include a plain-language reason.

为每个需要B-roll的节拍，根据以下维度对候选素材进行排名：

语义匹配度
支持角色适配度
证明力度
节奏适配度
视觉清晰度
可用片段质量

每个候选素材都应附带直白的理由说明。

4. Package the match result

4. 打包匹配结果

Build

broll-plan.json

with:

beat metadata
B-roll need decision
candidate list
fallback
keyword overlay suggestions
motion hints

构建

broll-plan.json

，包含：

节拍元数据
B-roll需求决策
候选素材列表
备选方案
关键词叠加建议
动态提示

Minimal Script

最小化脚本

First implementation script:

```
scripts/run_match_broll_plan.mjs
```

Current first-version behavior:

reads
```
chunked-<mode>.json
```
reads
```
broll-catalog.json
```
treats subtitle chunks as beat units
infers
```
beatRole
```
and
```
visualNeed
```
from spoken text
ranks B-roll candidates using conservative heuristics over:
- spoken keywords
- catalog
```
semanticTags
```
- catalog
```
supportRoles
```
writes
```
broll-plan.json
```

Example:

bash

node ${CLAUDE_SKILL_DIR}/scripts/run_match_broll_plan.mjs \
  --chunks /abs/path/chunked-basic.json \
  --catalog /abs/path/broll-catalog.json \
  --output <work-folder>/broll-plan.json

首个实现脚本：

```
scripts/run_match_broll_plan.mjs
```

当前第一版本功能：

读取
```
chunked-<mode>.json
```
读取
```
broll-catalog.json
```
将字幕分段视为节拍单元
从口播文本推断
```
beatRole
```
和
```
visualNeed
```
使用保守启发式算法对B-roll候选素材进行排名，考量维度包括：
- 口播关键词
- 素材库
```
semanticTags
```
- 素材库
```
supportRoles
```
写入
```
broll-plan.json
```

示例：

bash

node ${CLAUDE_SKILL_DIR}/scripts/run_match_broll_plan.mjs \
  --chunks /abs/path/chunked-basic.json \
  --catalog /abs/path/broll-catalog.json \
  --output <work-folder>/broll-plan.json

Release-Shell Execution Contract

发布壳执行协议

keep beat chunks, candidate rankings, and intermediate match plans under
```
<work-folder>/.postplus/broll-match-engine/
```
keep only final user-facing
```
broll-plan.json
```
or review exports outside
```
.postplus/
```
start with a bounded first pass on one transcript plus one catalog before broader edit planning
if required upstream inputs are missing or invalid, stop immediately instead of switching to ad hoc shell glue

将节拍分段、候选素材排名及中间匹配方案存储在
```
<work-folder>/.postplus/broll-match-engine/
```
目录下
仅将最终面向用户的
```
broll-plan.json
```
或审核导出文件放在
```
.postplus/
```
目录外
在进行更广泛的剪辑规划前，先针对一份文稿和一个素材库完成有限的首轮处理
如果所需上游输入缺失或无效，立即停止运行，而非临时拼凑脚本逻辑

Output Standard

输出标准

A strong output should let a human editor answer:

where to cut away
what to cut to
how long to stay there
what the fallback is if the preferred clip is weak

If the output cannot guide a real edit, it is not specific enough.

优质输出应能让人工剪辑师明确：

何时切换画面
切换到什么素材
停留时长
若首选素材质量不佳，备选方案是什么

如果输出无法指导实际剪辑，则说明不够具体。

First-Version Boundary

第一版本边界

Prefer:

explainable candidate ranking
beat-level JSON
optional markdown review

Do not require:

fully automatic timeline assembly
NLE export
guaranteed perfect semantic matching
motion graphics rendering

优先实现：

可解释的候选素材排名
节拍级JSON输出
可选的markdown审核文档

无需实现：

全自动时间线组装
NLE导出
完美语义匹配保障
动态图形渲染

Review Rule

审核规则

Always distinguish:

essential proof
useful support
optional filler

If the inventory does not contain strong proof footage, the plan should say so directly.

始终区分：

核心证明素材
有用辅助素材
可选填充素材

如果素材库中没有高质量的证明素材，方案应直接说明这一点。