Video Studio (read first; overrides the rest of this file)

This skill is installed in Video Studio (Edits/.claude/skills/), from the NextPhase / creator packs Bryan downloaded on 2026-10-31. Video Studio's CLAUDE.md wins wherever the two disagree.

• Start from the studio job, never an intake. Jobs come from the studio page (jobs/<id>/job.json). This skill runs when the job's style is floating-window. Never ask "short or long?" or "which style?": the page already asked. (nextphase-studio's intake is not used here.)
• One transcript: projects/<id>/transcript/words.json (WhisperX). Don't run this skill's own transcriber.
• Brand: styles/brand.json (purple #8B5CF6 / #C4B5FD / #4C1D95, Inter + Playfair Display) replaces this skill's default colours. Generate its theme/tokens from it.
• Show the motion graphics before rendering (Bryan's rule): design first, show stills of each graphic over the real frame in batches (the design → graphics stages, graphics.json), wait for approval, then render.
• Sounds and music: the studio's library (styles/sfx.json, node studio/sfx.mjs place) on the locked picture; music only from tracks Bryan owns or licensed.
• Windows Git Bash, not macOS/WSL: python not python3; a venv's Python is Scripts/python.exe; long paths fail, so map a short drive (cmd //c "subst R: <winpath>"). Run this skill's setup/doctor script only the first time the style is used, and ask before any download over ~500 MB.
• Where work goes: editing work in projects/<id>/, deliverables in jobs/<id>/.

ZAK — editing style

A talking head goes in; an edit comes out where the framing is the design. There is no grid, no stage, no collage, no accent colour and no motion. There is a black field, one rounded window sitting in it with real air above and below, a serif line in that air, and words that change three times a second.

⚠ Before editing ANY video, ask the user which style: zak (this skill) · persa · nextphase (nextphase-long-editing) · motiongraphics · vibecode. Never assume, never infer it from the footage. Record the answer at the top of PLAN.md.

Where this look comes from

Reverse-engineered frame by frame from one reel Jade supplied: REF-A — instagram.com/p/Db2RzkxvWpz/ — 1080×1920, 30fps, 39.9s, creator zacharyronusac.

Every number in assets/tokens.css marked MEASURED was read off it, and the tokens were verified by rendering them back through headless Chrome and diffing against the source frame — the window, the title and the caption all land within 1px (reference/stills/09-reference-vs-tokens.png).

⚠ It is one reel. Geometry and type are trustworthy to the pixel. The grade was measured in one room under one lamp and is a target range, not a calibrated curve. The scene vocabulary past the single locked frame is invented, and reference/layouts.md marks every scene measured or derived. reference/reference-log.md has the method, the raw readings, and a table mapping each finding to where it is written up.

⚠ That reel is another creator's work. Zak is the technique. Their words and their room are theirs — scripts/check-zak.sh fails the render if their copy resurfaces.

---

⛔ Seven hard rules

1. Never hardcode a colour, font or duration. Import assets/tokens.css; use its variables only. Never mix zak tokens with another style's.
2. The caption HARD-CUTS. Zero frames. No fade, no scale, no pop, no per-word reveal. 53 out of 53 measured changes landed on a single frame.
3. The caption has NO shadow, NO stroke and NO box. That is measured, not an omission — see reference/captions.md §2. It reads only because the plate under it is dark, so when it is hard to read, fix the plate. python3 scripts/check-contrast.py gates this on every render.
4. Nothing moves. No drift, no push-in, no parallax, no eased anything. Measured at 0px over 40 seconds. The one exception is the 6-frame full bleed, used once or twice in a whole video.
5. A caption is 2–4 words. A one-word caption is persa, not zak.
6. Every word on screen is derived from THIS video. Kit copy in components/ is placeholder and the reference copy belongs to another creator. Run bash scripts/check-zak.sh videos/<name> before every render.
7. Nothing publishes without explicit approval. Deliver rendered + scored, open it for review, stop there.

Read reference/ before building — captions.md (the core of the style — read this first) · layouts.md (the frame, the scenes, and the 16:9 derivation) · look.md · typography.md · motion-grammar.md · sound-design.md · workflow.md · reference-log.md.

---

1 · The look in one paragraph

A pure-black field with one rounded window in it, equal air above and below, and a single serif line in the upper air that is set once and never changes. Inside the window: warm, low-key footage that never touches white, and two to four white words in a brutally tight grotesque that replace themselves every twenty-odd frames. Two colours — black and white. Two typefaces. One move in the entire video. Everything else is a hard cut carried by a sound.

---

2 · The frame

1920×1080 @30fps is the default for this studio. The style was measured at 1080×1920, so :root in tokens.css carries those numbers and .zk-16x9 overrides the five that do not transfer.

<div class="zk-frame zk-16x9"> … </div>   <!-- long-form  -->
<div class="zk-frame">         … </div>   <!-- vertical   -->

	9:16 — measured	16:9 — derived
Window	968 × 968, a square	1720 × 732, a 2.35:1 scope letterbox
Side margin	56px (5.19%)	100px (5.19%) — the same
Air above / below	476 / 476, equal to the pixel	174 / 174, equal
Corner radius	70px	70px — the same absolute size
Title line	Playfair 56px, 66px above the window	Playfair 46px, 21px above
Caption	Inter 700, 63px, −0.065em	Inter 700, 80px, −0.065em

Read reference/layouts.md §5 before building wide. The 16:9 window is a genuine re-crop of the source, not a scale — zak-plate.sh takes a crop-y because a talking head wants headroom thrown away, not chin. And the 16:9 caption size is the one token in the style that was chosen by eye rather than derived; the reasoning, and why arithmetic gives two answers 3× apart, is written up there and in the tokens file.

⭐ Three things follow from the ground being literally #000000

• No drop shadow on the window. There is nothing behind it to catch one.
• No border, no rim, no glow. The radius is the only edge treatment.
• Nothing lives in the bottom air. It is empty for all 40 seconds of the reference. No bar, no handle, no logo. The air is the design.

⭐ .zk-frame on the composition root

Type is sized in cqh so one scale serves both aspects, and cqh resolves against a size container. Without .zk-frame every string renders at a fraction of its size and the layout still looks plausible — which is why it is the first thing to check when a scene looks subtly wrong. Never size type in %: that resolves against the parent's font-size and fails the same way.

---

3 · The scenes

Scene 1 is measured. Scenes 2–5 are derived — invented so long-form has a rotation, keeping everything that was measured intact. Full spec in reference/layouts.md §3.

#	Scene	For
1	Locked Window	the rest beat — and it is most of the runtime
2	Media Window	the window holds the real resource instead of the face
3	Split Window	the face and the thing, at once, one outer radius
4	Title Card	the chapter break — empty window, serif line carries it
5	Full Bleed	the one move. 6 frames. Once or twice per video.

Never the same scene twice in a row. Never hold one past ~20s. But note what the reference does: it holds one shot for 39.9 seconds. This style is supposed to sit still, and cutting every three seconds is the amateur read of it. Hold 6–15s at a stretch.

⭐ REAL MEDIA, NOT TYPESET TEXT

Every beat that supports a point shows a real resource — a screen recording, real product UI, a docs or pricing page, a repo card. ≥1 per 20s, minimum 8.

• Check every screenshot actually rendered content. Login-walled and JS-heavy pages come back blank and ship as empty windows.
• No random humans as b-roll. Cut to a person only when that person is the subject of the line.
• Never use emojis. There is no accent colour here to build a mark from — a designed mark is white type, a white rule, or nothing.
• Re-encode every sourced clip to 30fps CFR and verify with ffprobe.

⭐ COPY IS DERIVED, NEVER INHERITED

Every string comes from the speaker's own words at their onset in transcript-tight.json, the real thing named in that beat, or a verified figure. Write the actual copy into PLAN.md, per beat, before authoring.

---

4 · The two things that make it zak

The caption system

2–4 words, white, Inter 700 at −0.065em, hard-cutting on a single frame every ~22 frames, in a fixed position, with no shadow of any kind.

python3 scripts/build-captions.py assets/transcript-tight.json \
    --out seg-01-cold-open/captions.html --offset <segment start>
python3 scripts/check-contrast.py out/seg-01.mp4

The builder derives its hold target from the transcript's own speech rate — the measured 22 frames is what 2–4 words cost at the reference speaker's ~240 wpm, not a metronome. It reports rate, holds and words-per-caption against the reference and warns on anything out of range.

The checker is the one that matters. The caption has no shadow, so the plate is the only thing making it readable; a bright plate kills it silently, and the render still succeeds. It is calibrated: the reference reads 125.1 against a ceiling of 150, and a brightened copy of the same reel fails on every frame. reference/captions.md §2.

The stillness

Nothing drifts. Nothing pushes in. Nothing eases. Measured at 0px of drift and 0% of scale change across 40 seconds, with the title line static to within JPEG noise. This is the exact opposite of motiongraphics, and carrying that skill's continuous-drift habit over is the most common way a zak comp goes wrong. check-zak.sh fails a tween on .zk-plate or a transform on .zk-window.

---

5 · Motion

reference/motion-grammar.md. The whole table:

Move	Frames @30fps
Caption change	0 — it hard-cuts
Window contents change	0
Scene change	0
Camera drift / push-in	0 — there is none
Window → full bleed	6 — the only animation in the style

The cut is the transition. If a cut needs help, the help is a sound.

⚠ GSAP fromTo defaults to immediateRender:true and will apply a later beat's from state at frame 0. Always pass immediateRender:false — with a single tween in the whole piece, a failure here is total.

---

6 · Sound

reference/sound-design.md. Zak has no visual transitions at all, so sound is doing the job a transition does elsewhere — it is what makes a hard cut read as intentional. ~6–10 cues/min: transitions only, quiet transients.

Never cue a caption change. At 80 changes a minute it is unlistenable, and it is the most tempting mistake in the style.

Reuse the sibling palette and call the mixer by its path inside that skill folder — never copy it into a video folder:

bash ../nextphase-long-editing/scripts/mix-sfx.sh out/visual.mp4 out/sfx.mp4 sfx-cues.txt
bash ../nextphase-long-editing/scripts/mix-bgm.sh  out/sfx.mp4    out/final.mp4 assets/music/bed.mp3

---

7 · The workflow

Full detail in reference/workflow.md. Prefix every bash step with:

export PATH="$HOME/.nvm/versions/node/v24.18.0/bin:$HOME/.local/bin:$PATH"

0. Ask which style. Locate videos/<name>/, VIDEO.md, assets/recording.mp4.
1. Transcribe → chunk anything over ~90s, gap-check minute by minute.
2. Tighten → transcript-tight.json is now the only timing source.
3. Reframe + grade the plate (zak-plate.sh). No matte step — zak does not cut the subject out; reaching for rvm-matte.py here is persa habit.
4. PLAN.md — scenes per beat, the real on-screen copy, the media list. Get a "go" before authoring.
5. Source real media, verify facts, normalise to 30fps CFR.
6. Author per chapter. Symlink assets into each seg-NN/; HyperFrames rejects ../assets/. Build captions with build-captions.py.
7. Per segment: lint (0 errors) → check-zak.sh (exit 0) → snapshot → read the contact sheet and fix → render -q high -f 30 → check-contrast.py (exit 0).
8. Concat via ../nextphase-long-editing/scripts/concat-segments.sh — chapter timestamps come from real rendered durations, never from PLAN.md. Relay the source audio after the concat (AAC priming drifts ~38ms per join).
9. Sound: SFX pass, then the ducked bed.
10. Deliver: out/final.mp4 · out/chapters.txt · out/<name>.srt. Report rendered + scored, open for review. Publish nothing without approval.

---

Files in this skill

• assets/tokens.css — the measured design tokens and every zak class. Import into every composition. (CSS comments do not nest — never write a closing marker inside one of its header blocks, or the whole sheet silently stops applying.)
• components/zak-system.html — the whole style on one page: all five scenes at real proportions in both aspects, the type table, the cadence table, and a list of what breaks it. Open this first; it is the fastest way to see what the style is. Open it beside reference/stills/.
• reference/ — captions.md · layouts.md · look.md · typography.md · motion-grammar.md · sound-design.md · workflow.md · reference-log.md.
• reference/stills/ — the source evidence: full frames, a 1fps contact sheet, 3× type crops, a 4× corner crop, an 8-frame strip across a caption hard cut, the tokens-vs-reference round-trip, and the derived 16:9 form.
• scripts/zak-plate.sh — reframe a source into the window and grade it.
• scripts/build-captions.py — transcript → hard-cut 2–4 word phrase captions.
• scripts/check-contrast.py — the gate that matters. Run on every render; the caption has no shadow and dies silently on a bright plate.
• scripts/check-zak.sh — the static render gate. Run before every render.
• Everything else is shared with nextphase-long-editing — tighten, mix-sfx, mix-bgm, concat-segments. Call them by their path in that folder; never copy them into a video folder.