agentleFS
alisadiq-aialisadiq-ai/favstash-ai-video-editing-skills/skills/3-styles-on-camera/breakout-card/SKILL.md

breakout-card

Skill75 starsChanged 40 days agoApache-2.0View on GitHub

Style for edit-video: a white grid page with the speaker in a rounded card along the bottom and their head breaking out above it, dark pills and cards that blur in and out, and one-to-two-word Montserrat captions, cut against full-camera beats. Use for talking-head explainers that alternate graphics with personal lines, or when the creator asks for the head pop-out card.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Write the chunks from the verified transcript with ASR timing: one or two words, about 0.3–0.7s each, breaking on punctuation, lowercase …

    line 39
  2. Very specific:

    Write the run's build script with layouts, captions and one scene per page section (example_build.py shows mechanics only):

    line 89
  3. Very specific:

    Check and render with .favstash-studio/runtime/node_modules/.bin/hyperframes.

    line 96
  4. Distinctive:

    If frame 0 renders black, clone frame 1 over it before assembly.

    line 97

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Breakout card

A style on top of edit-video, which owns the cut, sound and review. The speaker sits in a rounded card along the bottom of a white grid page, and their head (and raised hands) break out above the card's top edge. Above them, one dark pill or card at a time explains the line, and one or two words of caption sit between the graphic and the head. Full-camera beats cut in for personal lines.

The look (tested defaults)

ElementSetting
Page#fcfcfd with a static 54px grid in #eaeaef
Cardx 60–1020, top 1395, 18px radius, bleeding off the bottom; the camera's full width fills it
BreakoutThe card's top edge cuts just above the eyes or glasses, so hair or a cap rise about 150–180px above it
GraphicsDark material #1c1c1f, white text, one soft shadow, a faint inner rim. Pills 96px tall, cards 34–36px radius, at y 200–1000
TypeMontserrat (OFL, bundled in assets/fonts). Pills 38px Bold, card text 32–37px SemiBold, headline 112px ExtraBold + 150px Black with grey-gradient ink
CaptionsOne or two words, lowercase as spoken, Montserrat Bold 60px. Page: dark ink between the graphic and the head. Camera: white, above the head
AccentNavy for tags, lines and badges. Green and red only for a real check or a real strike

Tokens sit in scripts/breakout.py (TOKENS); component CSS in assets/breakout-card.css.

Layouts and rhythm

  • Page for every informational beat. Cam (full frame) for personal lines and the name drop. Camz adds a slow push (1.2→1.27) for a re-hook like "here's the catch".
  • Alternate every 4–10s with hard cuts, placed on the cut's jump cuts where possible so the jump hides inside the layout change.
  • Frame 1 is the page with real proof already moving, such as a tilted capture flying in.
  • On a camera beat, a big white title can replace the caption for the name drop.

Captions

Write the chunks from the verified transcript with ASR timing: one or two words, about 0.3–0.7s each, breaking on punctuation, lowercase except file names and the CTA keyword. They are baked into the render because the font and word rhythm are the look; the compiler writes captions.json so a wording fix is a quick re-render. A chunk crossing a layout change splits and restyles automatically. Check the longest chunk in both layouts against edit-video's safe area.

Graphics vocabulary

ComponentUse
shotReal capture as a 3D-tilted card, optional highlight sweep. drop=True for frame 1
counter_pillA spoken figure counting up to the real value
headlineSmall line, then a huge and a mega word
tile_dropA file drops into a folder tile
checklistCard whose items appear and tick on their words
pillOne idea in a capsule; ok_at adds a check, strike_at a red strike and cross
cursor_clickHand cursor clicks a pill
tree + tree_wander / tree_straightAn agent wandering through folders vs. going straight to the right one
doc_card, chipA file that fills as it is described, then shrinks to a chip
commentCTA: a viewer comment types the keyword, chips pop for what they get
titleBig white name drop over the camera

Everything blurs into focus (opacity 0, scale .7–.86, blur 16px → sharp in about .35s) and blurs away (.25s). One graphic group at a time; it enters on the noun and acts on the verb. Real captures for facts; mockups stay generic and unmeasured meters say illustrative.

Head breakout plates

scripts/build_plates.py turns the cut camera plate (1080×1920, silent is fine) into a page plate and a full-camera plate. Needs Python 3.10+, numpy and opencv-python.

  1. prep <camera.mp4> <plate-dir> converts to a constant frame rate and makes the person matte: Apple Vision on macOS (compiled from person-matte.swift, needs the Xcode command line tools). Elsewhere, pass --matte <video> with a white-on-black person matte from any segmentation tool.
  2. still <plate-dir> 3 12 22 30 40 and inspect the head edge in each still. Tune with --set (saved to plates.json): cutRow (source row on the card's top edge), bgKey (removes wall-coloured fringe at the silhouette edge), objectKeys (a wall item right behind the head that the matte grabs) and protect (keeps keys away from the face and glasses).
  3. render <plate-dir> writes page-plate.mp4 and cam-plate.mp4.

Build

Needs the workspace runtime from edit-video's init-workspace.mjs --install.

  1. Cut the dialogue with edit-video; keep the word timeline.
  2. Make the plates (above) and capture proof.
  3. Write the run's build script with layouts, captions and one scene per page section (example_build.py shows mechanics only):
    sys.path.insert(0, '<this skill>/scripts')
    from breakout import Reel
    reel = Reel('<run>/work/breakout-v01', page_plate='<plates>/page-plate.mp4', cam_plate='<plates>/cam-plate.mp4', total=1363)
  4. Check and render with .favstash-studio/runtime/node_modules/.bin/hyperframes. If frame 0 renders black, clone frame 1 over it before assembly.
  5. Mix dialogue and SFX in your editor, or scripts/preview_mix.py <cues.json> <dialogue.wav> <out.wav> for a quick review mix. Review with edit-video.

Sound map (edit-video's SFX pack)

Visible eventCue
Frame 1hook-impact
Page ↔ cameracamera-shutter 4 frames early; the light shutter into a page
Re-hook pushmetallic-riser, resolving on the cut back to the page
Graphic arrivesui-pop (rows and chips quieter)
Tick, click, count landingui-click
Blur-away, line draws, file dropszoom-swish; sub-impact where something lands
Earned yesclean-ding.wav, once

Pitfalls already paid for

  • Matte the constant-frame-rate copy that prep makes; a matte from a variable-rate cut drifts a frame at every gap.
  • A coloured light on hair or a cap can make it as tinted as the wall behind it, only darker. The sampled-wall key with the default thresholds keeps it; a hue-only key eats it.
  • Keys inside the silhouette punch holes in reflective glasses; protect stops that.
  • Measure long headline words: 196px overflowed a 1080 canvas; the mega line is 150px.
  • Draw-on paths need their measured length (smooth_path returns it).
  • Animate the element inside a .row, never the row itself (it is the centring wrapper).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.