Agent Skills
› Forward-Future/scarlet
› b-roll
b-roll
GitHub用于视频编辑流程中B-roll和图形的规划与制作。先以文字描述覆盖点并生成待办,由编辑确认;随后仅执行已批准的素材查找或图形生成任务,确保画面配合语音节奏。
Trigger Scenarios
需要为视频内容添加B-roll或图形说明时
编辑要求生成特定的B-roll素材或动态图形时
Install
npx skills add Forward-Future/scarlet --skill b-roll -g -y
SKILL.md
Frontmatter
{
"name": "b-roll",
"category": "Picture and graphics",
"description": "Plan B-roll and graphics as written descriptions first, then find or make only the ones the editor asks for."
}
B-roll and graphics
Cover what's being said with something the viewer can see. Plan in words first; the editor decides what gets made.
First, describe
- When asked for B-roll or graphics, find the spots and leave a To do at each one. Make nothing, find nothing and render nothing at this stage.
- Each To do says what should be on screen and whether it's B-roll or a graphic, in a sentence the editor can judge without seeing it: "B-roll: screen recording of the agent running a task, start to finish." Name the line it covers and roughly how long to hold it.
- A spot earns a To do when a picture would show what's said: the thing named, the screen being described, a figure worth seeing, a jump cut that needs covering. Leave out anything that is only decoration.
- About one spot every 30 seconds at most, unless the editor asks for more. Fewer, stronger ideas beat a full list.
Then, make the ones asked for
- When the editor sends a To do back asking for it, make that one and only that one.
- For B-roll, use footage in the project's Media. If nothing there fits, say what to shoot or find, and leave the To do in place.
- For a graphic, follow /motion-graphics.
- Keep the speaker's voice running under B-roll. Never cover the face on the line that carries the point.
- Hold B-roll about 2 to 5 seconds and come in and out on a beat, not mid-word.
Version History
- 6e93a9e Current 2026-11-19 15:04


