Agent Skills › Forward-Future/scarlet › frame-screen

frame-screen

GitHub

指导屏幕录制时的画面构图与变焦逻辑。聚焦当前操作主体,过滤无关界面元素;通过平滑缩放突出关键内容(如代码或文本);根据语音指令调整视角,保持画面稳定并避免微小抖动。

skills/frame-screen/SKILL.md Forward-Future/scarlet

Trigger Scenarios

生成屏幕录制视频 处理演示类媒体内容 优化视频剪辑中的镜头运动

Install

npx skills add Forward-Future/scarlet --skill frame-screen -g -y
More Options

Use without installing

npx skills use Forward-Future/scarlet@frame-screen

指定 Agent (Claude Code)

npx skills add Forward-Future/scarlet --skill frame-screen -a claude-code -g -y

安装 repo 全部 skill

npx skills add Forward-Future/scarlet --all -g -y

预览 repo 内 skill

npx skills add Forward-Future/scarlet --list

SKILL.md

Frontmatter
{
    "name": "frame-screen",
    "category": "Picture and graphics",
    "description": "Frame a screen recording tight on what the speaker is doing or showing, with smooth zooms in and out as the subject changes."
}

Frame the screen on what I'm doing

While the screen recording is on picture, the frame holds what I'm doing or showing, and nothing else. The subject is the pane or panel it happens in, whole, with the title, labels and controls that belong to it: not a detail inside it, not the window around it, and not the desktop around that.

  • When I play, use or demo something, that pane fills the frame: the game, the preview, the canvas. Sidebars, chat panels, file lists, toolbars and the desktop around it stay out, even when they are part of the same window.
  • Whenever text is on screen for the viewer to read, because I read it aloud, talk about it or show it (a post, a reply, a headline, a line of code, a screenshot or page shown over my camera), zoom in on it: the passage fills the frame, large enough to read along at the video's size, and the frame follows it down the page.
  • When I point at one thing (a chart, a post, a result, a line of code), that thing fills the frame, whole.

What surrounds the subject stays out of the video: the desktop, the taskbar or dock, browser tabs and the address bar, other windows, a call, notifications. They come in only when they are what I'm showing.

What I'm saying and where the action is on screen decide the subject. When it moves, for example from reading the agent's reply to playing what it built, the frame moves with it. It holds still through small pointer moves and only changes when the subject does.

Zooming in and out

Every change of subject is a smooth zoom, never a jump. When I start referring to something, push in on it: the move starts just as I mention it and settles within about a second, easing in and out. When I move on, pull back out to the wider view, or go straight to the next subject if I name one right away, with the same easing. A zoom out is as smooth as a zoom in.

Hold each frame long enough to take in, at least a couple of seconds. If I touch on things faster than that, stay on a frame wide enough to hold them all instead of moving back and forth. The moves follow what I say, not the pointer. Zooming in on text keeps to the same restraint: one push in for a passage, not one per line or sentence. Following it down the page is a gentle move at the same size, and when I glance at several bits of text close together, hold one readable frame on all of them rather than hopping between them. Moving from one row or line of the same chart, table or page to the next isn't a change of subject: while what I read stays in view, hold the frame. A move that only shows a little more or less of what is already in view is one the viewer barely notices, so don't make it.

Before you set a new frame on the screen, look at the one already there. If the new one shows mostly the same thing, a little larger, smaller or shifted, keep the frame you have: a small reframe reads as a twitch, not a choice. A cut in the recording isn't a reason to reframe either: a jump cut on the screen doesn't need a new framing to hide it. The save report gives each move's zoom and shift, and a move of a few percent on the same subject usually means the frame should have held. Reframe only when what I'm on changes, or when the new frame shows something the viewer needs and the old one doesn't. This is a guide, not a rule: make the small move when it clearly helps.

Go tight by default, and no tighter than the subject: the whole pane, card, passage or chart I'm on stays in view. Push in on one part of it only when I single that part out, and come back out when I move on. A frame is tight enough when the text I'm reading is easy to read at the video's size; tighter than that only loses what surrounds the detail and softens the picture. Widen only when the viewer needs to see two things together, such as a prompt and its result side by side. The picture always fills the video, with no empty background beside it. It is the recording as it was recorded, moved and zoomed: don't rebuild the screen from pieces of it laid side by side. When the subject is taller or wider than the video's frame, show the part of it I'm using instead of letting what surrounds it in. Look at the framed video across the whole stretch, including mid-zoom, so the frame doesn't cut off a HUD, a control or a line I use a moment later.

My camera stays in its corner at the same size throughout: the screen is framed around it, not the camera moved around the screen. It never covers what I'm talking about: every line I read or refer to, every part of a chart or page I discuss, and every control I use stays fully visible. So place the subject clear of the camera's corner. Moving the camera is the last resort: only for a stretch you have framed and looked at with the camera in place, where no framing keeps the subject clear of it. Then move it for that stretch alone, bring it back afterwards, and say in that edit's reason what its corner would have covered. Other text on screen may sit under my camera, but always try to keep it clear too, with a framing that still shows the subject as it should. When I speak straight to camera without referring to anything on screen for a stretch, consider showing my camera full screen instead of the screen recording.

Switching between the screen and my camera

A switch between my camera full screen and the screen, whether the screen recording, a screenshot or a page shown over my camera, is a bigger change than a zoom: the viewer has to find their place again. So switch sparingly, and let each side hold for a stretch, not a moment. While I'm still talking about what the screen shows, the screen stays up: setting it up, saying what I asked for, going through what came back, reacting to it and judging it are all talking about it. Lean toward leaving the screen on: my camera comes back full screen once I've clearly moved on to something the screen doesn't show, not at the first sentence that could go either way. A short cut to my camera to land one strong point, then straight back to the screen, can work now and then, but while I'm on the screen, the screen is the default. What reads badly is switching often, or a screen or camera stretch so short the viewer barely takes it in before it changes again. When the screen would only be up for a glance, leave it up longer, or fold it into a longer stretch of screen nearby, rather than flashing it in and out. A few frames of camera between two stretches of screen, or of screen between two of camera, reads as a glitch: keep the screen on through it, or line the switch up with the cut beside it. Watch the run of switches as it plays before keeping it.

screen_reading says where I read aloud from the screen and screen_motion where the picture changes; with timeline_transcript, screen_text, media_frame, screen_area and timeline_framing, they cover what this needs.

Version History

  • b4fc0e1 Current 2026-11-22 10:51

    移除了关于保持画面的技能引用指导;新增在跳转剪辑时保持帧画面的逻辑,并通过百分比比较判断是否需重新构图。

  • dbfdef6 2026-11-22 04:41

    优化摄像头位置逻辑,改为固定摄像头并围绕其框选屏幕;增强屏幕画面的识别与保留机制,确保说话人提及屏幕内容时画面持续显示。

  • 27a27c9 2026-11-21 16:18

    新增记录镜头覆盖区域信息(如遮挡文本);优化在同一图表内阅读时的帧保持逻辑,避免无效微缩放。

  • f4e426c 2026-11-20 03:21

    修正了关于屏幕显示时长的描述,明确在演讲者仍在讨论相关内容时保持屏幕画面,仅在话题转移时才允许切至摄像头。

  • 59df58f 2026-11-19 02:52

    优化了文本缩放逻辑:当屏幕上出现或朗读文本时,直接对整段文本进行缩放以增强可读性,避免逐行频繁缩放导致的视觉跳跃。

  • b6bdb58 2026-11-18 20:45

Same Skill Collection

skills/add-captions/SKILL.md
skills/b-roll/SKILL.md
skills/dialogue-pass/SKILL.md
skills/fit-length/SKILL.md
skills/make-shorts/SKILL.md
skills/matt-talking-head/SKILL.md
skills/motion-graphics/SKILL.md
skills/punch-in/SKILL.md
skills/speed-through-waits/SKILL.md

Metadata

Files
0
Version
b4fc0e1
Hash
d35722a6
Indexed
2026-11-18 20:45

ホーム - Wiki
Copyright © 2011-2026 iteam. Current version is 2.156.0. UTC+08:00, 2026-11-23 00:52
浙ICP备14020137号-1