@autoaiai/aicut 0.1.11 → 0.1.13

package/skills/hyperframes-motion-director/references/hyperframes-stability.md

@@ -1,99 +0,0 @@
-# HyperFrames Stability
-
-HyperFrames rendering is frame-based and should be deterministic. A composition that previews once is not necessarily safe to render. Build timelines so any frame can be captured independently and consistently.
-
-## Deterministic Contract
-
-For render-critical visuals:
-
-- Use fixed width, height, fps, and duration.
-- Use local assets or assets guaranteed to exist before rendering.
-- Use seeded randomness if randomness is needed.
-- Use explicit timeline times.
-- Build timelines synchronously before rendering starts.
-- Keep media timing controlled by HyperFrames/runtime expectations.
-
-## GSAP Pattern
-
-Use paused timelines:
-
-```js
-const tl = gsap.timeline({ paused: true });
-
-tl.from(".title", {
-  y: 80,
-  opacity: 0,
-  duration: 0.8,
-  ease: "expo.out"
-}, 0.2);
-
-window.__timelines = window.__timelines || {};
-window.__timelines["root"] = tl;
-```
-
-Prefer absolute position parameters so timing can be reviewed and adjusted:
-
-```js
-tl.to(".product", { scale: 1, opacity: 1, duration: 0.7 }, 2.4);
-tl.to(".stat", { y: 0, opacity: 1, duration: 0.5 }, 5.8);
-```
-
-## Avoid
-
-Do not use these for main animation timing:
-
-- `Date.now()`
-- Unseeded `Math.random()`
-- `setTimeout`
-- `setInterval`
-- `requestAnimationFrame`
-- Asynchronous timeline creation
-- Runtime network fetches for required assets
-- Infinite repeats such as `repeat: -1`
-- Manual `video.play()` or uncontrolled `currentTime` changes
-
-These can make frame capture inconsistent or unavailable when rendering frames out of real-time order.
-
-## Asset Discipline
-
-Before rendering:
-
-- Confirm required images, videos, audio, and fonts exist locally or are vendored.
-- Avoid depending on remote fonts at render time.
-- Use stable file names.
-- Prefer explicit dimensions for images/video containers.
-- Include fallbacks only when they preserve layout.
-
-## Validation Commands
-
-Use the installed CLI's syntax. Start with:
-
-```bash
-npx hyperframes doctor
-npx hyperframes lint
-npx hyperframes validate
-npx hyperframes inspect
-npx hyperframes snapshot <composition> --at <times>
-```
-
-If a command fails because of missing dependencies, report the blocker and continue with checks that can run.
-
-## Snapshot Strategy
-
-Capture:
-
-- First frame.
-- Each scene's hero frame.
-- Each transition midpoint.
-- CTA lockup.
-- Any frame where dense text or product imagery appears.
-
-Snapshots catch layout and readability issues earlier than full review renders.
-
-## Common Failure Diagnosis
-
-- Preview works but render differs: check time-dependent code and asynchronous timeline setup.
-- Text cut off: inspect bounding boxes, safe margins, fixed container sizes, responsive font rules.
-- Missing image or font: check asset paths and render-time network calls.
-- Transition cuts to empty frame: remove premature exit animation or move transition earlier.
-- Motion feels random: replace implicit timing with explicit positions and beat-map timestamps.

package/skills/hyperframes-motion-director/references/imagegen-asset-pipeline.md

@@ -1,227 +0,0 @@
-# Image Gen Asset Pipeline
-
-Use this reference for premium vertical productions that need generated scene backgrounds, foreground objects, component sheets, transparent cutouts, or an image asset manifest.
-
-## Core Principle
-
-Count meaning before counting images.
-
-The asset set is derived from:
-
-1. Distinct visual worlds.
-2. Source-specific objects that must move independently.
-3. Product or story claims that need visible proof.
-4. Transition objects that physically hand attention between beats.
-5. Resolution and isolation needs.
-
-Do not begin from a target such as four backgrounds or twelve components. Those numbers may be correct for one film and wrong for the next.
-
-## Phase 1 Asset Analysis
-
-### Visual worlds
-
-Group beats only when they can share all of these:
-
-- The same spatial metaphor or product environment.
-- The same lighting and camera logic.
-- A compatible text-safe zone.
-- The same story job.
-- A transition that does not need a new reveal world.
-
-Split a new visual world when:
-
-- Hook, mechanism, proof, confidence, or CTA needs a different spatial meaning.
-- Reusing the stage would make the film look like title cards over wallpaper.
-- The subject position or quiet text zone changes materially.
-- The transition needs a new object, opening, chamber, route, surface, or destination.
-
-Premium multi-scene work normally needs multiple independent vertical backgrounds. The exact count equals the approved visual-world analysis, not the duration.
-
-For every world, record:
-
-- ID.
-- Source beats.
-- Why it cannot share another background.
-- Background role.
-- Focal subject.
-- Quiet text zone.
-- Target scenes.
-- Transition entry and exit.
-
-### Movable foreground inventory
-
-List a foreground object only when it must:
-
-- Replace explanation.
-- Move independently.
-- Prove a source claim.
-- Carry attention into another beat.
-- Change state.
-- Create the main memory hook.
-
-For every object, record:
-
-- Exact source phrase.
-- Story role.
-- Target scene.
-- Motion action.
-- Expected on-screen scale.
-- Whether it is a hero object.
-- Deletion test.
-
-The accepted list decides the transparent cutout count.
-
-## Sheet Strategy
-
-Premium vertical generated-image work uses at least one source-driven component sheet. A separately generated hero can supplement the sheet, but cannot replace it. If the entire foreground must remain official, supplied, or code-native and no Image Gen component is truthful, record a user-approved pipeline exception instead of silently setting the sheet count to zero.
-
-Use one sheet when:
-
-- Every object remains large enough to crop cleanly.
-- Objects fit with generous separation.
-- One perspective and lighting direction suit the whole inventory.
-- No hero object needs a large on-screen scale.
-
-Use multiple sheets when:
-
-- The full inventory would crowd one sheet.
-- Different component families need different perspective or material.
-- Small marks and large product surfaces would waste each other's space.
-
-Generate an object separately when:
-
-- It is the hero.
-- It fills a large part of the 1080×1920 frame.
-- Fine material or edge detail matters.
-- The sheet crop would need enlargement beyond its native resolution.
-
-Sheet count is a resolution and isolation decision, not a style score.
-
-Keep commands, screenshots, labels, UI states, and proof text official or code-native. Put only bitmap-suitable, product-specific objects, material shells, symbols, stamps, transition pieces, and other visually useful foreground elements on the generated sheet.
-
-## Component Sheet Prompt Contract
-
-Every component-sheet request states:
-
-- Exact inventory and grid position.
-- One coherent palette, material, border, shadow, perspective, and lighting direction.
-- Large non-overlapping isolation space around every item.
-- Flat transparent or declared matte background.
-- No shared floor, connected shadow, overlap, or merged glow between cells.
-- No final screen copy, fake UI text, fake logos, watermarks, labels, or decorative filler.
-- One object per cell.
-- Hero objects excluded when they need separate generation.
-
-Do not ask Image Gen to label the cells. Map positions in `ASSET_MANIFEST.json`.
-
-## Background Prompt Contract
-
-Generate backgrounds independently at a native vertical ratio.
-
-Every background request states:
-
-- Visual-world ID and story job.
-- 9:16 target.
-- Focal subject and crop-safe region.
-- Quiet text zone and platform-safe boundaries.
-- Shared palette, material, lens, lighting, and contrast.
-- What visually changes from the other worlds.
-- No baked-in copy, logos, fake UI, watermarks, or random icons.
-
-Accept the first strong background as a style anchor. Use it to keep later backgrounds and component sheets in the same visual world.
-
-## Cutout Contract
-
-Preserve source sheets under `assets/images/source/`.
-
-For every accepted item:
-
-1. Crop its cell without neighboring objects.
-2. Remove the transparent or declared matte background.
-3. Trim unused space without touching the object.
-4. Add transparent padding.
-5. Save a named PNG under `assets/images/components/`.
-6. Check that transparency is real, not only an alpha channel with opaque pixels.
-7. Check that every outer edge is fully transparent.
-8. Check for green, blue, white, or black matte spill.
-9. Check native resolution against expected display scale.
-10. Record the path and source cell in the manifest.
-
-Build two proof sheets:
-
-- Dark background: catches light halos and pale matte residue.
-- Light background: catches dark halos, lost shadows, and black matte residue.
-
-Reject a cutout when:
-
-- The object touches an outer edge.
-- The alpha channel is fully opaque.
-- Matte color remains in visible pixels.
-- Another component leaks into the crop.
-- Fine detail collapses at intended scale.
-- The object needs a label to explain its role.
-
-## Asset Manifest
-
-Use `templates/ASSET_MANIFEST.template.json`.
-
-The manifest records:
-
-- Production decision, user-approved exception status, and exception reason.
-- Background, component, sheet, and separate-hero count reasons.
-- Visual-world inventory.
-- Background IDs, paths, roles, scenes, and quiet zones.
-- Component inventory.
-- Sheet strategy plus every sheet ID, path, grid, transparent/matte background mode, matte color when applicable, and accepted ID.
-- Cutout paths, generation method, source sheet/cell when applicable, source phrases, target scenes, motion actions, and hero status.
-- Dark/light proof sheets.
-- Revision history whenever the approved asset analysis changes.
-
-The approved analysis owns the numbers. Phase 2 fills accepted local paths. If the count changes, update the reason and record the revision before animation.
-
-Required item shapes:
-
-```text
-visual_world: id, source_beats[], reason, background_id
-background: id, path, scene_ids[], role, quiet_zone
-component_sheet: id, path, grid, background_mode, matte_color, accepted_component_ids[]
-component: id, path, generation_method, source_sheet_id, source_cell, source_phrase, role, target_scenes[], motion_action, hero
-revision: timestamp, field, previous_value, new_value, reason, approval_reference
-```
-
-Use `background_mode: "matte"` with a six-digit `matte_color`, or `background_mode: "transparent"` with `matte_color: null`. Use `generation_method: "sheet"` for cropped sheet items and `"separate"` only for supplemental hero generations. Keep `revisions` as an empty array until an approved decision changes.
-
-## Validation
-
-Run:
-
-```bash
-node scripts/validate_image_assets.mjs <project-dir>
-node scripts/check_assets.mjs <project-dir> --strict --require-premium-assets
-```
-
-The validator checks:
-
-- Declared worlds and generated backgrounds agree.
-- Declared component inventory and transparent cutouts agree.
-- Sheet strategy agrees with the number of sheets and any separately generated objects.
-- Backgrounds are independent and approximately 9:16.
-- PNG files are decodable.
-- Cutouts contain real transparency and transparent outer edges.
-- Declared matte color is removed.
-- Proof sheets exist.
-- Composition source uses backgrounds and cutouts.
-- Composition source does not use the source sheet directly.
-
-Validation proves asset integrity, not directing quality. Static hero frames and rendered video stills still decide whether the assets produce a strong film.
-
-## Review Questions
-
-- Does every background represent a genuinely different visual world?
-- Could two backgrounds be merged without losing meaning? If yes, merge them.
-- Does any beat reuse a background only to save work? If yes, regenerate.
-- Does every cutout move, prove, transform, or hand off attention?
-- Is the sheet crowded because of an arbitrary item target?
-- Does a hero object need a separate high-resolution generation?
-- Do the dark and light proof sheets show clean edges?
-- Would the frame fall back to black background plus text if the cutouts were removed?

package/skills/hyperframes-motion-director/references/motion-background-system.md

@@ -1,165 +0,0 @@
-# Motion And Background System
-
-Use this when planning generated images, text-over-image layouts, GSAP choreography, or visual review.
-
-## Why Videos Look Weak
-
-Most weak motion videos fail for the same reasons:
-
-- Text floats on a flat or generic background.
-- The background reads as wallpaper instead of a stage for the idea.
-- Motion adds activity but does not guide attention.
-- Too many elements compete for hero status.
-- Paused hold frames lack a clear message.
-- Important text is placed in a safe lower area even when it should dominate the first attention zone.
-- Motion is only a slide entrance, so the video feels like a PPT export.
-
-The fix: background creates the world, support assets add semantic depth, typography carries the message, and motion directs the eye.
-
-Before generating a background or animating type over it, choose the layout contract from `references/text-over-background-layout.md`. The contract defines where the subject sits, where text can sit, which image ratio to generate, where mobile safe boundaries live, and where motion is allowed to travel.
-
-## Default Image Rule
-
-Do not assume a numeric image count. For new generated-image work, derive the background set from distinct visual worlds and write any exception in the brief:
-
-- Strong supplied product, UI, logo, or footage assets already carry the scene.
-- The direction is intentionally pure kinetic typography.
-- The task is a narrow edit of an existing composition.
-
-Premium multi-scene work requires multiple independent vertical backgrounds unless a user-approved exception applies. In Phase 1, analyze distinct visual worlds and movable foreground objects. After confirmation, use Codex Image Gen by default for needed bitmap assets and move accepted project-bound files into `assets/images/`. Use `references/imagegen-asset-pipeline.md` for the manifest, sheet, cutout, and validation workflow.
-
-## Image Count
-
-Do not count images from duration alone.
-
-- Background count equals the approved distinct visual-world count. Premium multi-scene work normally needs multiple independent vertical backgrounds.
-- Transparent component count equals the accepted source-driven movable-object inventory.
-- Sheet count depends on usable resolution, isolation space, component families, and hero-object scale.
-- Official product/UI assets remain the first choice when they prove the claim more faithfully than generated assets.
-
-If two backgrounds do the same job, keep the stronger one. If one sheet crowds the inventory, split the sheet or generate hero objects separately instead of shrinking everything.
-
-## Image Roles
-
-Each image gets exactly one primary role:
-
-- Stage: atmosphere, depth, and light.
-- Symbol: the central metaphor.
-- Texture: grain, haze, material, or light falloff.
-- Anchor: product, person, logo, UI, or real object.
-- Transition plate: mask, wipe, focus pull, or morph source.
-
-Reject images that only fill space.
-
-## Support Asset Decision
-
-Support assets are not optional decoration. They are the answer to whether the video needs more visual layers than the background stage. The answer can be "none", but it must be deliberate.
-
-Use support assets for:
-
-- Symbol or anchor: the recurring object, product surface, or visual motif the viewer should remember.
-- Texture or atmosphere: grain, haze, material, shadow plate, dust, or light falloff that makes the scene tactile.
-- Transition plate: matte, aperture, wipe source, mask, focus layer, morph source, or proof-safe cutout.
-- Semantic glyph: one minimal meaning-bearing mark, not an icon set.
-- Product / UI fragment: a faithful product, interface, chart, or proof crop used when the story needs evidence.
-- Motion accent: line, glint, trace, path, bracket, rail, or particle-like detail with a defined attention or transition job.
-
-Generate support bitmap assets only when image generation produces something HyperFrames cannot do better. Prefer CSS/SVG/HyperFrames for lines, masks, marks, glyphs, geometric accents, UI outlines, and anything that must stay sharp, recolorable, or tightly synchronized.
-
-Every support asset must specify:
-
-- Role and source: generated / supplied / code-generated / mixed.
-- Local path after generation when bitmap-based.
-- Shared style lock: palette, lighting direction, texture language, contrast, and perspective.
-- Safe zones: which text or platform regions it must avoid.
-- Motion purpose: reveal, depth, transition, proof, attention guide, or hold support.
-- Deletion trigger: what makes it noise.
-
-Never use a support asset only because the frame feels empty. First ask whether the copy, crop, scale, light, or timing is weak.
-
-## Text Over Image
-
-- Reserve a quiet text zone before generating or cropping.
-- Declare the layout contract, including text rectangle, subject rectangle, image ratio, title tier, mobile safe zones, and motion bounds.
-- Declare the first eye target. Hook text and amplified keywords usually own center or upper-center attention when the visual subject does not need that area.
-- Keep the focal subject away from title and CTA zones.
-- Use local treatment behind text: vignette, shadow plate, gradient mask, blur, or desaturation.
-- Put small proof text only on calm image areas.
-- If the text needs a card to survive, regenerate, recrop, or simplify the image.
-- If the subject is outside its crop-safe zone or text would fall into the bottom platform overlay zone, regenerate or recrop before implementation.
-
-## Motion Rules
-
-Default arc:
-
-```text
-World appears -> Title lands -> Metaphor reveals -> Proof locks -> CTA holds
-```
-
-Motion budget:
-
-- One primary motion idea per scene.
-- One optional support motion for atmosphere or transition.
-- Background motion stays slow: push, parallax, focus pull, or light sweep.
-- Background motion must preserve the quiet text zone through the full readable hold.
-- Text settles before it must be read.
-- Text motion must stay inside its declared motion bounds and avoid bottom/right/top mobile UI zones.
-- Important reveals need stillness after impact.
-- Multi-scene transitions should overlap or share an anchor; avoid empty black gaps.
-
-Prefer mask reveal, clip reveal, rim-light sweep, focus pull, parallax, cross-dissolve, scale settle, or morph-to-logo.
-
-Avoid repeated `y + opacity`, linear primary reveals, infinite loops, decorative particles, random icons, and moving text during reading time.
-
-## CSS3 / SVG / GSAP Motion Craft
-
-Use HTML/CSS/SVG/GSAP as a film language and rendering method.
-
-- CSS3: use `clip-path`, `mask-image`, `filter`, `mix-blend-mode`, perspective transforms, CSS variables, and local darkening when they express reveal, focus, pressure, or transition.
-- SVG: use paths, masks, strokes, scan lines, underlines, frames, connectors, and logo lockups when vector structure can guide attention.
-- GSAP: build a paused master timeline with labels and explicit positions. Use nested scene timelines only when they remain attached to the master timeline.
-- Kinetic typography: define entry, lock, emphasis, exit, and bridge for every important text block.
-- Text transitions should transform attention instead of only swapping lines.
-
-If a beat can be described as "text fades up over image," it is under-directed. Rewrite the motion in terms of mask, light, path, camera, crop, split, compression, assembly, or handoff.
-
-## Codex Image Gen Requirements
-
-Every generated-image plan should state:
-
-- Role and aspect ratio.
-- Scene/beat where the asset will be used.
-- Layout contract and image ratio.
-- Focal subject position.
-- Quiet text zone.
-- Text rectangle and subject rectangle.
-- Safe bottom/right/top platform boundaries for overlaid text.
-- Required negative space and crop-safe area.
-- Lighting direction and contrast.
-- Palette and forbidden colors.
-- No baked-in text unless required.
-- No fake logos, labels, explanatory icons, diagrams, watermark, or decorative clutter.
-- For support assets: transparent/cutout intent if needed, isolation margin, consistent lighting, and no unrequested icon set.
-- For asset sets: visual-world count reason, component inventory, sheet strategy, hero-object decision, manifest path, and dark/light cutout proof.
-- Local output path after acceptance.
-
-After generation, inspect the image before writing animation code:
-
-- Accept only images with a clear role, stable focal subject, usable text-safe space, and a standard ratio.
-- Recrop or regenerate when the quiet zone contains busy detail, bright seams, faces, product edges, fake UI, or accidental text.
-- Keep final text, proof notes, CTA, masks, parallax, focus pulls, and timing in HyperFrames unless the brief requires baked-in content.
-- Remove repeated assets that serve the same role.
-- Run `validate_image_assets.mjs` after cropping accepted component-sheet items and before animation.
-
-## Review Gate
-
-Reject the work if:
-
-- There is no background plan or explicit exception.
-- The background could be swapped without changing meaning.
-- Support assets are missing when the frame needs semantic depth, transition material, or product proof.
-- A support asset has no role, source, motion purpose, safe-zone rule, style lock, or deletion trigger.
-- Text readability depends on luck.
-- Motion can be removed without changing the story.
-- More than one layer tries to be the hero.
-- The hero frame is not pause-worthy.

package/skills/hyperframes-motion-director/references/motion-craft.md

@@ -1,182 +0,0 @@
-# Motion Craft
-
-Use this reference when a video risks feeling like a slide deck, static poster sequence, or text pasted over a background. The goal is directed motion: movement that changes what the viewer understands, where they look, and why they keep watching.
-
-## First Principles
-
-A motion video uses four moving systems:
-
-- Camera: where the viewer is in the space.
-- Light: what becomes important.
-- Type: how language arrives, transforms, exits, and leaves memory.
-- Transition: how one idea turns into the next without resetting attention.
-
-If only opacity and vertical position change, the work will feel like a PPT export. A frame can be still during its readable hold, but the video around that hold needs spatial movement, typographic behavior, and transition logic.
-
-## Anti-PPT Failure Modes
-
-Reject or revise when:
-
-- Every scene is a centered text block over the same background.
-- The only animation is `opacity` plus `y`.
-- Transitions are black fades or hard cuts with no shared visual anchor.
-- Text appears as completed lines instead of being revealed, assembled, masked, scanned, split, or transformed.
-- Background images are static wallpaper with no camera, focus, light, or mask behavior.
-- SVG/CSS shapes are absent even when a simple line, mask, frame, grid, path, or underline could carry the transition.
-- All beats have the same layout and rhythm.
-- The film can be represented as four screenshots without losing much.
-
-## Motion Vocabulary
-
-Use a small set of deliberate motion devices. Pick one signature device for the whole video, then vary it per beat.
-
-### Kinetic Text Relay
-
-Use this vocabulary when the film is driven by text, icons, and fast transitions. The goal is not to show more elements. The goal is to make every word feel acted upon.
-
-Define the chain before implementation:
-
-- Keyword: the word or short phrase that owns the beat.
-- Action object: the icon, cursor, waveform, timeline, frame, brush, scan rail, stamp, product tile, or line that gives the keyword a physical behavior.
-- Direction: left push, right push, upward lift, downward press, type-on, wipe, scan, split, compression, expansion, orbit, or hard cut.
-- Relay: the outgoing word must hand attention to the incoming word through the action object, direction, mask, line, camera move, or sound hit.
-- Hold: after the action lands, the word must become readable before the next hit.
-
-Strong examples:
-
-- A waveform grows through "Record" and pushes the word aside to reveal "Edit".
-- A timeline strip crosses the center line and becomes the mask that reveals the next title.
-- A cursor types the first letters, then the blinking caret becomes a vertical scan line.
-- A product tile slides upward, crops the old word, and locks as the frame for the next word.
-- A brush stroke reveals the word, then exits as a curved path that pulls in the next word.
-
-Weak examples:
-
-- Each word fades up in the same position.
-- Icons appear beside labels but do not move, mask, push, scan, or reveal anything.
-- A transition happens through empty black with no visual handoff.
-- Every beat uses the same direction, same center position, and same timing.
-
-### Camera And Space
-
-- Slow push into the subject.
-- Parallax between background, symbol, type, and dust/grain layers.
-- Rack focus or blur-to-sharp reveal.
-- 2.5D pan across a generated image or product surface.
-- Crop-window move that discovers the subject.
-
-### Light And Material
-
-- Rim-light sweep that reveals the metaphor.
-- Gold scan line that also becomes a mask, underline, divider, or CTA guide.
-- Volumetric haze drift behind the subject, never over readable text.
-- Specular glint on one meaningful edge.
-- Vignette or shadow plate that opens and closes with the message.
-
-### Kinetic Typography
-
-- Masked line reveal with `clip-path` instead of plain opacity.
-- Word-level stagger only for the keyword, not every character.
-- Split-line handoff where the outgoing word becomes the incoming word's anchor.
-- Highlight sweep on one amplified term.
-- Text assembly from fragments when the story is about synthesis.
-- Type compression/expansion when the story is about pressure or release.
-- Baseline slide or vertical wipe when the story is about progress or replacement.
-- SVG path text or curved underline only when the path carries meaning.
-
-### SVG And CSS3 Layers
-
-Use SVG/CSS when they can carry structure more crisply than bitmap assets:
-
-- SVG masks for wipes, reveals, circular scans, slash cuts, iris opens, and logo lockups.
-- SVG paths for motion trails, timeline arcs, underlines, map routes, scan curves, and proof connectors.
-- CSS `clip-path`, `mask-image`, `filter`, `mix-blend-mode`, `transform`, `perspective`, and `will-change` for deterministic visual depth.
-- CSS variables for timing, color, spacing, and safe-zone geometry.
-- Avoid complex SVG illustrations that become decorative diagrams.
-
-### GSAP Choreography
-
-Use GSAP as the timing engine for one coordinated sequence:
-
-- Build one paused master timeline.
-- Use labels for beats: `hook`, `tension`, `reveal`, `proof`, `cta`.
-- Use absolute positions or labels so edits are easy.
-- Register the timeline for HyperFrames control.
-- Animate transforms, opacity, filters, masks, and CSS variables; avoid layout-thrashing properties.
-- Keep text still during readable holds.
-- Use scene-level timelines only when they are nested into the master timeline.
-
-## Text Transition Rules
-
-Every beat with screen copy must define:
-
-- Entry: how the text arrives.
-- Lock: when it becomes still and readable.
-- Emphasis: which word or line receives the visual hit.
-- Exit: how it leaves without dragging through unsafe zones.
-- Bridge: what visual element connects it to the next beat.
-
-Bad: `title fades up`.
-
-Better: `title reveals through a left-to-right mask; README receives a gold scan highlight; the scan line exits as the next beat's divider`.
-
-For kinetic text relay promos, also define:
-
-- Old word exit direction.
-- New word entry direction.
-- Action object that bridges them.
-- Transition midpoint snapshot timestamp.
-- Whether fade is support or the main transition. If fade is the main transition, redesign unless deliberate stillness is the concept.
-
-## Short Vertical Hard Gates
-
-For promos 20 seconds or shorter:
-
-- One hook or core-viewpoint frame must use center or upper-center text impact unless a stronger visual subject clearly owns that zone.
-- The first 0-2 seconds must name the first eye target, biggest word/object, and scroll-stop motion event.
-- Three or more beats using the same textRect, same entry, and same rhythm is a PPT risk and should trigger storyboard revision.
-- At least one important text moment must use a real transition device: mask, scan, split, compression, assembly, path handoff, or an equally specific alternative.
-- A report that says "pass" while all important text lives in the lower half is too lenient unless every center zone is owned by a stronger subject.
-- If adjacent frames show only opacity/position changes, add stronger motion craft or explain why stillness is the concept.
-
-## Rhythm Shape
-
-For short vertical promos:
-
-- First readable hook should appear by 0.8-1.2s.
-- Use a strong movement before or during the hook, then stillness.
-- Alternate high-motion and low-motion beats.
-- Give the proof beat a crisp mechanical rhythm.
-- Give the CTA the calmest hold in the video.
-- Do not let all beats last the same visual length unless the repetition itself is the concept.
-
-## Review Gate
-
-Score the video harshly:
-
-- 0-30: static slides, decorative fades, no motion idea.
-- 31-55: readable but PPT-like; weak transitions and repeated entrances.
-- 56-75: coherent motion, but no memorable signature moment.
-- 76-90: strong typography, spatial movement, and meaningful transitions.
-- 91-100: one clear cinematic idea, one memorable motion device, every transition earns its place.
-
-Before delivery, inspect still frames and moving frames. A good still frame is necessary, but not sufficient. If the video loses little when reduced to screenshots, it needs stronger motion craft.
-
-## Kinetic Relay Scorecard
-
-Use this 100-point score when the requested result is a text/icon transition promo:
-
-- 20: first 0-2 seconds have a clear largest word/object and a real scroll-stop motion event.
-- 20: important words are revealed, pushed, typed, scanned, compressed, split, or otherwise acted upon.
-- 20: icons or objects participate in transitions instead of sitting as decoration.
-- 20: adjacent beats have relay continuity through direction, object, mask, line, cursor, scan, or camera movement.
-- 10: rhythm alternates between motion hits and readable holds.
-- 10: brand/CTA resolves cleanly and feels like the end of the chain.
-
-Interpretation:
-
-- 100: target state; ready for final render if snapshots and readability pass.
-- 90-99: usable draft only if the report names the missing points and the next edit.
-- 70-89: promising, but revise the weakest transition or rhythm beat.
-- 60-69: rebuild the transition map before render.
-- Below 60: still PPT-like; revise before delivery.