tutorial-screen
Screen-tutorial editor for software how-tos, coding walkthroughs and course demos: promise, prerequisites, numbered steps, recap; cuts the waiting, typing and loading, punches in on the area being clicked, adds step chips, keyboard-shortcut and callout overlays, blurs private dat
Install
npx skills add https://github.com/Viprasol-Tech/claude-video-editing-skills/tree/main/plugins/video-studio/skills/tutorial-screen
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install viprasol-tech-claude-video-editing-skills@llmmart
git clone https://github.com/Viprasol-Tech/claude-video-editing-skills.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole viprasol-tech/claude-video-editing-skills collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Tutorial Screen
A screen recording a viewer can follow without pausing: no waiting, the right thing magnified, every step labelled.
- Best for: software tutorials, coding walkthroughs, SaaS how-tos, course demos, bug-fix videos
- Inputs: a screen recording with your narration (best), or a screen recording + a facecam recorded at the same time; optional final project link
- Default skin:
chapter-pro· alternativestutorial-steps(facecam circle + step chips),terminal-tech,highlighter-study - Formats: 16:9 1920x1080 master, chapters,
captions.srt; optional 9:16 tip clips - Duration target: 5-15 min (one task per video); tips 30-60 s; genre preset
tutorial-screen
Before you start
- Needs video-studio-core in the same skills folder.
CORE=video-studio-core,SKILLS= the folder holding both. - First time:
python CORE/scripts/check_env.py. Preset:python CORE/scripts/editor.py genre tutorial-screen. - The
--genrepreset sets autocut's pause cut (0.45 s) and run.py respects it; pass--max-gaponly to override. - Ask what is private on screen (emails, API keys, customer names, tokens). It must be blurred before anything is published.
How a pro cuts this
| beat | % | what happens |
|---|---|---|
| promise | 0-5 | show the finished result first (5-10 s), then say what you'll build |
| prerequisites | 5-10 | versions, accounts, files needed (a checklist card) |
| numbered steps | 10-90 | one step = one chapter; say it, do it, show the result |
| recap | 90-97 | the steps again in 15 s |
| next / CTA | 97-100 | link, next video |
- Pacing: ASL 3-8 s; cut every wait (loading, installs, typing long strings - show a 1 s speed-up or skip to the result). Something changes every 4-6 s: a zoom, a chip, a callout.
- Cut rules: never cut away the click that matters; keep the result on screen >= 1 s after each action. Don't cut mid-sentence of an explanation.
- Zoom: zoom in when the UI text is small or the action is in one corner (1.6-2.5x on the area), zoom out before the cursor travels across the screen. Smooth 0.4-0.6 s moves, not jumps.
- Transitions: cuts; a step-chip change marks a new step. No whooshes on every click.
- Music/SFX: quiet bed (28 LU under) or none; soft
click/ui-taponly on the key clicks. - Colour: none - accurate UI colours (
"color": {"auto": false}). - Captions: sidecar SRT (YouTube) or
cleanpreset low on screen; keyboard shortcuts as their own chips. - Client judges: can a viewer follow every step without pausing or hunting for the cursor.
- Common mistakes: 4K UI shown full-frame (unreadable), cursor lost, leaked secrets, dead air while something installs, steps not labelled, music louder than the narration.
Make the edit
A. Narration recorded with the screen (one file):
python CORE/scripts/run.py screen.mp4 --style SKILLS/chapter-pro --genre tutorial-screen \
--smart-cuts --retakes --no-matte --model small --prompt "product names, commands" --work tut --stop-after plan
B. Screen + separate facecam recorded together: combine them into one picture-in-picture source first, so every cut removes the same moment from both (otherwise autocut desyncs the face from the screen):
ffmpeg -i screen.mp4 -i face.mp4 -filter_complex "[1:v]scale=360:-2[f];[0:v]scale=1920:-2[s];[s][f]overlay=W-w-40:H-h-40" \
-map 1:a -c:v libx264 -crf 18 -c:a aac -shortest tut_src.mp4
python CORE/scripts/run.py tut_src.mp4 --style SKILLS/chapter-pro --genre tutorial-screen --smart-cuts \
--retakes --model small --work tut --stop-after plan
(With tutorial-steps instead: run.py face.mp4 --clip screen.mp4 --no-cut ... shows the screen full-frame with the face in a circle, but only stays in sync without autocut.)
Then:
python CORE/scripts/scenes.py tut/cut.mp4 --work tut --threshold 0.15 # big screen changes = step boundaries
python CORE/scripts/editor.py review tut/plan.json --genre tutorial-screen --fix
python CORE/scripts/render.py tut/plan.json --frames 2,20,60,120 # LOOK: UI text readable?
python CORE/scripts/render.py tut/plan.json
python CORE/scripts/finish.py tut/plan.json --platform youtube
Blur private data with the plan's blurs list (box in output pixels, top-left corner; read the position off a
--frames still): "blurs": [{"x": 880, "y": 96, "w": 420, "h": 40, "start": 0, "end": 95.0}], or "pixelate": true
for a mosaic. It blurs the recording under the graphics, so chips and captions stay sharp. A secret that moves (scrolling,
a zoom) needs one box per time range; the engine has no tracking blur. See CORE/references/plan-schema.md.
Refine the plan
- Zoom on the action:
speaker.focusX/focusYset where a zoom aims for the whole video; per-moment zooms arespeaker.zooms({"start": 42.1, "end": 47.0, "scale": 1.8, "cx": 0.78, "cy": 0.08}:cx/cy= the clicked spot, 0-1 of the source frame; the engine does not follow the cursor by itself). When different zooms need different areas, put the screen in aclipelement and animate it with keys:"keys": {"s": [[0, 1], [0.5, 2.0, "inOut"], [4.5, 2.0], [5, 1, "inOut"]], "tx": [[0, 0], [0.5, -420]], "ty": [[0, 0], [0.5, 260]]}. - Step chips:
{"type": "chip", "text": "Step 2 · Add the API key", "start": 61.0, "end": 140.0, "x": 50, "y": 60, "anchor": "top-left"}and top-level"chapters": [{"start": 0, "title": "Result"}, {"start": 61.0, "title": "Add the API key"}]. - Prerequisites:
{"type": "checklist", "title": "You need", "items": [{"text": "Node 20", "at": 22.4}, {"text": "A free account", "at": 24.0}]}. - Shortcuts:
{"type": "tile", "text": "Ctrl + Shift + P", "start": 80.2, "end": 82.4, "x": 960, "y": 960}. - Callouts / pointer:
{"type": "callout", "point": [1510, 88], "x": 1200, "y": 300, "title": "Deploy button"},{"type": "pointer", "kind": "arrow", "x": 1510, "y": 88, "from": [1300, 400]}. - Code:
{"type": "codeblock", "title": "app.js", "code": "...", "diff": true}to restate code the viewer must copy (exact text from the recording). - Speed-up the wait: remove it with
--cut "start-end"(source seconds) on run.py, or keep 1 s with achip"2 minutes later". - Auto elements: the skin may add counters (labels built from nearby words, e.g. "% faster Follow") and a
subscribebutton that can collide with captions near the end. Fix the label text or delete them. - Sound:
"audio": {"voice": {"denoise": "light", "eq": "clear"}},"sfx": [{"t": 81.0, "kind": "click", "gain": 0.3}]. - Colour:
"color": {"auto": false},"finish": {"grain": 0, "vignette": false, "sharpen": 0.3}.
Deliver
python CORE/scripts/deliver.py tut/plan.json --to youtube --out-dir tut/deliver
python CORE/scripts/qc.py tut/deliver/final_youtube.mp4 --plan tut/plan.json --platform youtube
- Upload
captions.srt; pastechapters.txt(one chapter per step) into the description with the links. - Tips for Shorts: cut single steps with
retention-reel(--reframewill not help on UI; render a 9:16 plan with the screen in aclipelement zoomed on the area instead).
Check before delivering
- Watch every step at 1x: could you follow it without pausing? UI text >= ~24 px tall at 1080p when it matters.
- No secrets anywhere (scrub the contact sheet and the full video; also thumbnails).
- Results stay on screen >= 1 s after each action; no cursor-lost moments.
- Chapters match the steps;
qc.pypasses (-14 LUFS, -1 dBTP). - Tell the user: seconds removed, steps and chapter times, what was blurred, file paths.
Example requests
- "Edit my 25-minute screen recording into a clean tutorial, cut all the loading and zoom on the clicks."
- "Coding tutorial: add step numbers, chapters and show my shortcuts on screen."
- "My Loom with facecam - make it look professional and blur my API key."
Files (claude-video-editing-skills)
-
SKILL.md 8.4 KB
--- name: tutorial-screen description: "Screen-tutorial editor for software how-tos, coding walkthroughs and course demos: promise, prerequisites, numbered steps, recap; cuts the waiting, typing and loading, punches in on the area being clicked, adds step chips, keyboard-shortcut and callout overlays, blurs private data, and delivers a 16:9 video with YouTube chapters and captions. Use when the user says: edit my screen recording, tutorial video, how-to video, coding tutorial, software walkthrough, cut the loading screens, zoom on my cursor, add step numbers, make my Loom look professional. Default skin chapter-pro; any Video Studio skin works." license: MIT --- # Tutorial Screen A screen recording a viewer can follow without pausing: no waiting, the right thing magnified, every step labelled. - **Best for:** software tutorials, coding walkthroughs, SaaS how-tos, course demos, bug-fix videos - **Inputs:** a screen recording with your narration (best), or a screen recording + a facecam recorded at the same time; optional final project link - **Default skin:** `chapter-pro` · alternatives `tutorial-steps` (facecam circle + step chips), `terminal-tech`, `highlighter-study` - **Formats:** 16:9 1920x1080 master, chapters, `captions.srt`; optional 9:16 tip clips - **Duration target:** 5-15 min (one task per video); tips 30-60 s; genre preset `tutorial-screen` ## Before you start - Needs **video-studio-core** in the same skills folder. `CORE` = `video-studio-core`, `SKILLS` = the folder holding both. - First time: `python CORE/scripts/check_env.py`. Preset: `python CORE/scripts/editor.py genre tutorial-screen`. - The `--genre` preset sets autocut's pause cut (0.45 s) and run.py respects it; pass `--max-gap` only to override. - Ask what is private on screen (emails, API keys, customer names, tokens). It must be blurred **before** anything is published. ## How a pro cuts this | beat | % | what happens | | --- | --- | --- | | promise | 0-5 | show the finished result first (5-10 s), then say what you'll build | | prerequisites | 5-10 | versions, accounts, files needed (a checklist card) | | numbered steps | 10-90 | one step = one chapter; say it, do it, show the result | | recap | 90-97 | the steps again in 15 s | | next / CTA | 97-100 | link, next video | - **Pacing:** ASL 3-8 s; cut every wait (loading, installs, typing long strings - show a 1 s speed-up or skip to the result). Something changes every 4-6 s: a zoom, a chip, a callout. - **Cut rules:** never cut away the click that matters; keep the result on screen >= 1 s after each action. Don't cut mid-sentence of an explanation. - **Zoom:** zoom in when the UI text is small or the action is in one corner (1.6-2.5x on the area), zoom out before the cursor travels across the screen. Smooth 0.4-0.6 s moves, not jumps. - **Transitions:** cuts; a step-chip change marks a new step. No whooshes on every click. - **Music/SFX:** quiet bed (28 LU under) or none; soft `click`/`ui-tap` only on the key clicks. - **Colour:** none - accurate UI colours (`"color": {"auto": false}`). - **Captions:** sidecar SRT (YouTube) or `clean` preset low on screen; keyboard shortcuts as their own chips. - **Client judges:** can a viewer follow every step without pausing or hunting for the cursor. - **Common mistakes:** 4K UI shown full-frame (unreadable), cursor lost, leaked secrets, dead air while something installs, steps not labelled, music louder than the narration. ## Make the edit **A. Narration recorded with the screen (one file):** ```bash python CORE/scripts/run.py screen.mp4 --style SKILLS/chapter-pro --genre tutorial-screen \ --smart-cuts --retakes --no-matte --model small --prompt "product names, commands" --work tut --stop-after plan ``` **B. Screen + separate facecam recorded together:** combine them into one picture-in-picture source first, so every cut removes the same moment from both (otherwise autocut desyncs the face from the screen): ```bash ffmpeg -i screen.mp4 -i face.mp4 -filter_complex "[1:v]scale=360:-2[f];[0:v]scale=1920:-2[s];[s][f]overlay=W-w-40:H-h-40" \ -map 1:a -c:v libx264 -crf 18 -c:a aac -shortest tut_src.mp4 python CORE/scripts/run.py tut_src.mp4 --style SKILLS/chapter-pro --genre tutorial-screen --smart-cuts \ --retakes --model small --work tut --stop-after plan ``` (With `tutorial-steps` instead: `run.py face.mp4 --clip screen.mp4 --no-cut ...` shows the screen full-frame with the face in a circle, but only stays in sync without autocut.) Then: ```bash python CORE/scripts/scenes.py tut/cut.mp4 --work tut --threshold 0.15 # big screen changes = step boundaries python CORE/scripts/editor.py review tut/plan.json --genre tutorial-screen --fix python CORE/scripts/render.py tut/plan.json --frames 2,20,60,120 # LOOK: UI text readable? python CORE/scripts/render.py tut/plan.json python CORE/scripts/finish.py tut/plan.json --platform youtube ``` **Blur private data** with the plan's `blurs` list (box in output pixels, top-left corner; read the position off a `--frames` still): `"blurs": [{"x": 880, "y": 96, "w": 420, "h": 40, "start": 0, "end": 95.0}]`, or `"pixelate": true` for a mosaic. It blurs the recording under the graphics, so chips and captions stay sharp. A secret that moves (scrolling, a zoom) needs one box per time range; the engine has no tracking blur. See `CORE/references/plan-schema.md`. ## Refine the plan - **Zoom on the action:** `speaker.focusX`/`focusY` set where a zoom aims for the whole video; per-moment zooms are `speaker.zooms` (`{"start": 42.1, "end": 47.0, "scale": 1.8, "cx": 0.78, "cy": 0.08}`: `cx`/`cy` = the clicked spot, 0-1 of the source frame; the engine does not follow the cursor by itself). When different zooms need different areas, put the screen in a `clip` element and animate it with keys: `"keys": {"s": [[0, 1], [0.5, 2.0, "inOut"], [4.5, 2.0], [5, 1, "inOut"]], "tx": [[0, 0], [0.5, -420]], "ty": [[0, 0], [0.5, 260]]}`. - **Step chips:** `{"type": "chip", "text": "Step 2 · Add the API key", "start": 61.0, "end": 140.0, "x": 50, "y": 60, "anchor": "top-left"}` and top-level `"chapters": [{"start": 0, "title": "Result"}, {"start": 61.0, "title": "Add the API key"}]`. - **Prerequisites:** `{"type": "checklist", "title": "You need", "items": [{"text": "Node 20", "at": 22.4}, {"text": "A free account", "at": 24.0}]}`. - **Shortcuts:** `{"type": "tile", "text": "Ctrl + Shift + P", "start": 80.2, "end": 82.4, "x": 960, "y": 960}`. - **Callouts / pointer:** `{"type": "callout", "point": [1510, 88], "x": 1200, "y": 300, "title": "Deploy button"}`, `{"type": "pointer", "kind": "arrow", "x": 1510, "y": 88, "from": [1300, 400]}`. - **Code:** `{"type": "codeblock", "title": "app.js", "code": "...", "diff": true}` to restate code the viewer must copy (exact text from the recording). - **Speed-up the wait:** remove it with `--cut "start-end"` (source seconds) on run.py, or keep 1 s with a `chip` "2 minutes later". - **Auto elements:** the skin may add counters (labels built from nearby words, e.g. "% faster Follow") and a `subscribe` button that can collide with captions near the end. Fix the label text or delete them. - **Sound:** `"audio": {"voice": {"denoise": "light", "eq": "clear"}}`, `"sfx": [{"t": 81.0, "kind": "click", "gain": 0.3}]`. - **Colour:** `"color": {"auto": false}`, `"finish": {"grain": 0, "vignette": false, "sharpen": 0.3}`. ## Deliver ```bash python CORE/scripts/deliver.py tut/plan.json --to youtube --out-dir tut/deliver python CORE/scripts/qc.py tut/deliver/final_youtube.mp4 --plan tut/plan.json --platform youtube ``` - Upload `captions.srt`; paste `chapters.txt` (one chapter per step) into the description with the links. - Tips for Shorts: cut single steps with `retention-reel` (`--reframe` will not help on UI; render a 9:16 plan with the screen in a `clip` element zoomed on the area instead). ## Check before delivering - Watch every step at 1x: could you follow it without pausing? UI text >= ~24 px tall at 1080p when it matters. - No secrets anywhere (scrub the contact sheet and the full video; also thumbnails). - Results stay on screen >= 1 s after each action; no cursor-lost moments. - Chapters match the steps; `qc.py` passes (-14 LUFS, -1 dBTP). - Tell the user: seconds removed, steps and chapter times, what was blurred, file paths. ## Example requests - "Edit my 25-minute screen recording into a clean tutorial, cut all the loading and zoom on the clicks." - "Coding tutorial: add step numbers, chapters and show my shortcuts on screen." - "My Loom with facecam - make it look professional and blur my API key."
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.