@autoaiai/aicut 0.1.11 → 0.1.13

package/skills/hyperframes-motion-director/references/visual-standard.md

@@ -1,174 +0,0 @@
-# Visual Standard
-
-Use this as the single visual reference for proposal, layout, motion, and review.
-
-## Style Lock
-
-- Background: `#050505`.
-- Palette: white, gray, restrained warm gold.
-- Mood: low-brightness cinematic, high contrast, large negative space.
-- Texture: subtle paper grain, shallow depth of field, volume haze, thin rim light, local metallic highlights.
-- Composition: one dominant symbol or phrase per frame.
-- Background imagery: default for new videos unless a supplied-asset or pure-code exception is documented. It must function as stage, symbol, texture, anchor, or transition plate.
-- Visual objects: use a small set of project-specific props, marks, frames, stamps, brackets, rails, or texture pieces only when they compress meaning, guide attention, prove a claim, or create a memory hook.
-
-Avoid ordinary illustration, ecommerce banner layout, icon piles, generic neon tech, multicolor palettes, busy collage, decorative particles, and explanatory diagrams.
-
-## Metaphor First
-
-Extract the underlying idea before choosing imagery:
-
-- Core viewpoint.
-- Largest conflict.
-- Emotional center.
-- Keyword to amplify.
-- Visual metaphor.
-
-The metaphor must be legible without labels. If the viewer needs explanatory icon labels to understand the frame, the frame is too literal or too noisy.
-
-## Source Text To Visuals
-
-Do not design from an asset pack first. Design from the source text, product mechanism, and proof claims first.
-
-Before implementation, write a visual lexicon:
-
-- Exact source phrases that deserve visual treatment.
-- Nouns/entities that can become objects, surfaces, chips, cards, marks, or stages.
-- Verbs/actions that can become motion, state changes, routes, stamps, locks, reveals, or handoffs.
-- Numbers/modules/metrics that can become proof moments.
-- One motif and three to five component families that repeat across scenes.
-- A rejection list for attractive but generic visuals.
-
-Large text can be a hook, but large text boxes cannot be the default substitute for visual design. Replace paragraphs with product surfaces, object systems, kinetic keywords, proof fragments, chips, or diagrams. If a still frame cannot reveal what source material inspired it without reading a paragraph, the frame is visually generic.
-
-## Visual Objects
-
-Visual objects replace words, direct the eye, or make the process believable.
-
-Use:
-
-- One primary object system per short video, such as source paper, film frame, product surface, lens, rail, scan path, artifact, map route, or timeline.
-- 1-3 functional mark types, such as a confirmation stamp, inspection bracket, proof tick, crop guide, scan rail, underline, or lockup mark.
-- Sparse texture pieces only when they belong to the metaphor, such as paper fragments for documents, dust for atmosphere, or light traces for scanning.
-
-Avoid:
-
-- Icon clusters, per-line icons, generic checkmarks, rockets, sparkles, lightning, AI chips, fake dashboards, and decorative badges.
-- Objects that require text labels to make sense.
-- Objects that duplicate the copy instead of reducing it.
-- Multiple object languages in one 20s vertical promo.
-
-Before implementation, run the removal test: if deleting the object changes nothing about comprehension, attention, trust, or memorability, remove it.
-
-## Support Assets
-
-Support assets are the production form of visual objects: generated bitmaps, supplied crops, SVG marks, code-generated masks, or motion accents used by the renderer. They may improve cinematic quality only when they remain subordinate to the dominant idea.
-
-Use support assets for:
-
-- A recurring symbol or anchor that carries the metaphor.
-- Texture, haze, light falloff, shadow plate, material detail, or proof-safe crop that makes the frame tactile.
-- Mask, matte, glint, trace, rail, path, or transition plate that gives motion a physical bridge.
-- Product/UI fragment, data object, or evidence crop when the beat needs proof.
-- Minimal semantic glyphs when they compress meaning without explaining the frame.
-
-Reject support assets when:
-
-- They fill empty space without changing meaning.
-- They form an icon pile, diagram, or second visual language.
-- They compete with title, symbol, product, or CTA.
-- Their lighting, texture, perspective, or contrast does not match the scene.
-- Their motion can be removed without changing attention, reveal, proof, or transition.
-- They enter `textRect`, bottom platform overlays, or other declared safe zones.
-
-## Typography And Layout
-
-- Default output format is Simplified Chinese vertical promotional video: 9:16, `1080x1920`, with platform-safe top and bottom zones.
-- For text over or near background imagery, choose a layout contract before image generation. Use `references/text-over-background-layout.md`.
-- Use one huge title, one quiet support line, and one CTA/brand lockup.
-- Declare the first eye target for every hero frame. Hook text, emotional keywords, and central claims usually own the center or upper-center attention zone. Use lower-half copy for CTA/proof holds or subject-dominant frames.
-- Default maximums: title 1-2 lines, support text 1-2 lines, CTA 1 line, one proof/stat cluster.
-- For Chinese copy, use fewer characters per beat, intentional line breaks, larger type, and stronger contrast than horizontal desktop layouts.
-- Compress user content into the smallest readable claim. A beat should carry one idea instead of a source paragraph.
-- Use fixed font sizes per breakpoint. Avoid viewport-width font scaling.
-- Default letter spacing is `0`; tracking is allowed only for small all-caps labels.
-- Set max width, max lines, line-height, and overflow behavior for every text block.
-- Long words, mixed Chinese/English copy, subtitles, and CTA labels must stay inside their containers.
-- Keep safe margins platform-specific: larger center-safe zones and stronger bottom protection for 9:16, calmer bottom spacing for 16:9.
-- Keep CTA, subtitles, proof notes, support lines, and brand lockups above the declared safe bottom boundary.
-- Keep small text away from right-side short-video controls and top platform UI unless the delivery surface is known to be clean.
-- Use no more than two type families unless brand assets require it. Avoid default-looking display typography when the title is the hero.
-
-## Background Images
-
-- The background must either carry meaning, create spatial depth, ground the product, or disappear behind the message.
-- The background must match a declared layout contract with text rectangle, subject rectangle, quiet text zone, and crop-safe subject zone.
-- Reserve a quiet text zone in the image crop. Type should not sit on high-frequency texture, bright edges, faces, product seams, or accidental tangents.
-- Use vignette, mask, blur, shadow plate, darkening, or desaturation locally behind text before considering a card.
-- Generated images should avoid baked-in text, fake UI, fake logos, labels, explanatory icons, watermarks, and random decoration unless specifically required.
-- Each Codex Image Gen asset needs one declared role, target scene, standard ratio, focal subject position, quiet text zone, crop-safe region, and accepted local path.
-- Inspect generated images before motion work. Regenerate or crop when busy detail, bright edges, faces, product seams, fake UI, or accidental text enters the quiet text zone.
-- More images are not better, and fixed counts are not direction. Generate one independent vertical background per declared visual world. Merge worlds that genuinely share space and story job; split them when reuse would turn the film into text over wallpaper.
-- Derive transparent foreground components from the accepted source-driven object inventory. Use multiple sheets or supplemental separate hero generations when one sheet would crowd the assets or reduce usable resolution.
-
-## Motion
-
-Motion should reveal meaning, not decorate the frame.
-
-Use slow push, subtle parallax, focus pull, rim-light reveal, mask wipe, clip reveal, SVG path motion, CSS masks, scan-line bridges, and kinetic typography. Avoid repeated `y + opacity` entrances, empty fades between scenes, restless motion, and decorative effects.
-
-Motion, background, type, and transition are one system. Background movement must preserve the quiet zone, text must land inside its rectangle before reading time, and transitions must bridge meaning through subject, axis, mask, light, or camera direction.
-
-Text motion must have a transition idea: entry, readable lock, emphasis, exit, and bridge. The most important word should not merely appear; it should be revealed, assembled, scanned, split, compressed, expanded, or handed off in a way that matches the story.
-
-For text/icon transition promos, use a kinetic relay standard:
-
-- The viewer remembers a keyword chain, not paragraphs.
-- Each major word gets one action object or visual behavior.
-- Icons and objects must push, mask, scan, type, crop, reveal, compress, split, or hand off the next word.
-- Direction changes should be designed: left/right push, up/down push, scan, wipe, type-on, crop, or hard cut with a visible relay.
-- Empty black fades are allowed only for deliberate silence or final release.
-- Every transition needs a midpoint that can be inspected as a designed frame.
-
-Typical timing:
-
-- Premium reveal: 0.8-1.6s, `expo.out` or `sine.inOut`.
-- Title hold: 0.6-1.2s after entrance.
-- Proof/stat hold: 0.8-1.5s.
-- CTA hold: 1.0s or more.
-
-Motion budget:
-
-- One primary motion idea per scene.
-- One optional support motion for atmosphere, focus, or transition.
-- Background motion should be slow and camera-like.
-- Text must become still before the viewer needs to read it.
-- One signature motion moment per video is enough; too many special elements weaken the frame.
-- A good video should lose something important when reduced to screenshots. If screenshots tell the whole story, upgrade the motion craft.
-
-## Review Gate
-
-Reject the work if:
-
-- It skips user confirmation after `BRIEF_DESIGN_PROPOSAL.md`.
-- It looks like a banner, infographic, icon cluster, or generic tech poster.
-- It has no background-image plan or explicit pure-code/supplied-asset reason.
-- It has no layout contract, text rectangle, subject rectangle, title tier, motion bounds, or mobile safe boundary for text-over-background frames.
-- The background image is interchangeable wallpaper, fights the text, or needs a card to rescue readability.
-- It has more than one dominant message in a frame.
-- The first eye target is unclear or the main hook is unnecessarily pushed into the lower half.
-- The first-frame or hero-frame dominant visual mass is accidentally off-center, clipped, or balanced only by hidden/offscreen elements.
-- Text overlaps, clips, shrinks unpredictably, or falls outside safe margins.
-- The work defaults to 16:9, English copy, or desktop-style density without a user or platform reason.
-- It is only background plus text when the concept needs a product, process, proof, or memory object.
-- It uses generic office, AI, productivity, dashboard, paper, or desk assets that are not mapped to source phrases, nouns, verbs, numbers, or product states.
-- It has many visual components but no shared motif, component family system, or source-text mapping.
-- Large paragraph cards or oversized text boxes carry beats that should have been converted into objects or proof surfaces.
-- It claims a visual asset system while `assets/images` has only placeholders, or every beat uses the same dark stage with text and translucent cards.
-- Added visual objects are generic decoration, not part of the story.
-- The first readable frame appears too late.
-- The metaphor is surface-level or needs labels.
-- All elements move at once, motion repeats the same entrance pattern, or important text never settles.
-- It feels like PPT: static cards, repeated fade-ups, lower-third explanations, empty black fades, and no text transition or scene bridge.
-- A text/icon transition promo has no keyword chain, no active action-object chain, no direction map, or no relay continuity between adjacent beats.
-- The video passes render checks but fails still-frame review.

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

@@ -1,320 +0,0 @@
-# Production Workflow
-
-This workflow turns a vague motion-video request into a reviewable HyperFrames production. Follow it in order for new videos. For edits, change only the affected stage.
-
-The default creative direction is a Chinese-first vertical promotional film with black-background cinematic metaphor. Extract the essence of the article, theme, product, or argument, then build a restrained symbol that carries the point.
-
-New videos need a content-derived background decision by default. Premium multi-scene work normally needs multiple independent vertical backgrounds, one for each declared visual world. It also needs a support-asset decision: none, generated, supplied, code-generated, or mixed. Backgrounds create space, carry metaphor, protect text readability, or ground the product. Support assets add semantic depth, tactile texture, product proof, or transition material when the background alone is too flat.
-
-For text over or near background imagery, use `references/text-over-background-layout.md`. Choose a candidate layout strategy before image generation, then lock the final per-beat layout contract after generated or supplied imagery is inspected. The final contract owns image ratio, text rectangle, subject rectangle, quiet text zone, title tier, mobile safe zones, and motion bounds.
-
-For motion craft, use `references/motion-craft.md`. Static hero frames are necessary. The finished piece must define where attention starts, how text moves, how scenes bridge, and which CSS/SVG/GSAP devices make it feel like motion rather than slides.
-
-For text/icon transition references, use kinetic text relay as the default structure. This means the film is built from a keyword chain, small action objects, directional pushes/wipes/scans, readable holds, and transition midpoint snapshots. The key question is not "what does each card say?" but "how does one word physically hand attention to the next?"
-
-New video work is two-phase: first produce a brief/design proposal and wait for user confirmation; only then produce images, HyperFrames source, snapshots, renders, and review artifacts.
-
-## Output Writing Standard
-
-Artifacts should read like production notes.
-
-- Use short, direct sentences.
-- State project-specific choices, constraints, and risks.
-- Keep user-facing proposal copy in plain Chinese by default.
-- Avoid repeated contrast-pivot phrasing in Chinese or English.
-- Avoid self-talk, internal reasoning, generic praise, inflated sales language, and unrelated commentary.
-- Cut paragraphs that do not help approve, produce, validate, or review the video.
-
-## 1. Intake
-
-Capture the minimum viable production brief:
-
-- Source article, theme, or product idea.
-- Goal: what should the viewer do after watching?
-- Audience: who is this for?
-- Language: default to Simplified Chinese promotional copy unless the user explicitly asks for another language.
-- Platform: default to vertical social video, such as Douyin / TikTok / Reels / Shorts, unless the prompt names a different surface.
-- Aspect ratio and resolution: default to 9:16 1080x1920. Use 16:9 or other sizes only when the user, platform, or delivery surface clearly requires it.
-- Duration: default to 10-15 seconds for cinematic promotional motion video or kinetic typography.
-- Product or offer: what is being promoted?
-- Proof: what can be truthfully claimed?
-- Style: default to black cinematic metaphor unless the user explicitly overrides it.
-- Assets: logo, screenshots, product footage, audio, fonts, brand colors.
-- Hard constraints: required text, forbidden claims, legal copy, language, accessibility.
-
-When details are missing, write assumptions instead of stalling. Ask only if the missing detail changes the whole production.
-
-## 1.5. Essence Extraction
-
-Before design, extract:
-
-- Core viewpoint.
-- Largest conflict.
-- Emotional center.
-- The keyword that deserves visual amplification.
-- A visual metaphor that can carry the video.
-
-Translate abstract meaning into one restrained visual symbol:
-
-- AI replacement -> erased human silhouette.
-- Anxiety -> a thin line about to snap.
-- Time -> countdown suspended in darkness.
-- Growth -> micro-light inside a crack.
-- Information overload -> data fragments pulled into a black void.
-- Long-termism -> a single distant lamp.
-- Platform migration -> black obelisk or data tower.
-- Automation -> documents entering a silent machine.
-
-Reject literal surface drawings unless the user explicitly asks for them.
-
-## 2. Brief Design Proposal
-
-Before any implementation, produce `BRIEF_DESIGN_PROPOSAL.md` or the equivalent response structure:
-
-- Essence and metaphor.
-- Structure choice.
-- Language, platform, aspect ratio, pixel size, duration, FPS, safe margins.
-- Writing style: plain, compact, project-specific, no self-talk, no generic pitch language, no repeated contrast-connector phrasing.
-- Image generation decision, including default background image role or the explicit reason it is not needed.
-- Asset count analysis: distinct visual worlds, movable foreground inventory, count reasons, sheet strategy, separately generated hero objects, and `ASSET_MANIFEST.json`.
-- Support asset plan: role, source, style lock, safe zones, motion purpose, and deletion trigger for any symbol, texture, anchor, transition plate, semantic glyph, product/UI fragment, or motion accent.
-- Typography, layout, overflow handling, and mobile crop handling.
-- Motion plan and risk gates.
-- Background text layout plan: candidate layout strategy, likely text/subject relationship, title tier, mobile safe-zone risks, and regenerate/recrop trigger.
-- Attention map: first eye target, center-impact decision, and rationale for center/upper-center/side/lower-safe text.
-- Motion craft plan: camera behavior, kinetic typography, text transition style, SVG/CSS structure, GSAP timeline structure, signature motion moment, and anti-PPT risk.
-- Kinetic text relay plan when relevant: keyword chain, action-object chain, directional transition map, relay object, readable holds, transition midpoint snapshots, target kinetic relay score of 100, and no-delivery threshold below 90.
-
-Stop here and ask for confirmation. Generate images and write composition code after the user confirms.
-
-## 3. Design System
-
-The design system prevents style drift.
-
-Define:
-
-- Background and foreground color. Default to `#050505`, white/gray, and restrained warm gold.
-- Accent color and whether any warm rim glow is allowed. Avoid colorful gradients by default.
-- Display and body typography.
-- Safe margins.
-- First eye target and center-impact policy for hook, central claim, proof, and CTA frames.
-- Layout grid or placement logic.
-- Background image system: role, crop, focal subject, quiet text zone, contrast treatment, and local asset path plan.
-- Support asset system: role, source, content-derived count decision, shared style lock, safe zones, motion purpose, code-generated versus bitmap-generated split, and deletion triggers.
-- Background text layout system: default layout strategies, allowed final contract variants, image ratios, text rectangles, subject rectangles, title tiers, mobile safe zones, motion bounds, and regenerate/recrop triggers.
-- Text-over-image rules: where type may sit, how the image will be darkened/blurred/desaturated behind text, and what crop is forbidden.
-- Component patterns such as title card, product frame, proof stat, and CTA.
-- Motion personality and motion budget.
-- Motion craft system: camera movement, kinetic typography, text transitions, SVG/CSS layers, GSAP labels, and scene bridges.
-- Kinetic text relay system when relevant: keywords, action objects, direction vocabulary, transition handoffs, midpoint snapshots, and scorecard target.
-- Do and don't rules.
-- The central metaphor symbol and how it is lit.
-- Any generated image assets needed before HyperFrames composition.
-
-If the user's product already has a brand system, respect it only where it does not destroy the requested cinematic metaphor. If not, use the strict house style and document the choice.
-
-## 4. Storyboard And Copy
-
-For short motion videos, script is compression. Put copy directly in the storyboard so text, timing, layout, and motion stay together.
-
-Default short motion structures:
-
-```text
-10s: Hook -> Tension -> Metaphor Reveal -> CTA
-15s: Hook -> Tension -> Metaphor Reveal -> Proof -> CTA
-30s: Pattern interrupt -> Conflict -> Metaphor Reveal -> Proof -> Outcome -> CTA
-```
-
-For no-voiceover videos, script means screen text and timing. Every beat should have one job.
-
-For kinetic text relay videos, script is a chain of memory words. Keep the copy short enough that each word can receive a designed action. Prefer 4-8 beats such as `Hook word -> product action -> proof word -> outcome word -> CTA` over explanatory sentences.
-
-Choose structure by material:
-
-- Center symbol for trends, insight, AI, platform, or philosophy.
-- Huge title for conflict, suspense, emotion, or sharp viewpoint.
-- Person anchor for tutorial, interview, personal brand, or methodology.
-- Huge number for growth, milestone, or data shock.
-
-## 5. Storyboard Detail
-
-The storyboard is where direction happens. Keep it explicit.
-
-Every beat should specify:
-
-- Start and end time.
-- Main message.
-- First eye target and whether the text should own center attention.
-- Hero frame timestamp.
-- Metaphor role.
-- Layout and hierarchy.
-- Layout contract, including textRect, subjectRect, safeBottomY, title tier, and motion bounds.
-- Motion.
-- Text transition: entry, readable lock, emphasis, exit, and bridge.
-- Kinetic relay details when relevant: old word exit, new word entry, action object, transition direction, relay handoff, and midpoint snapshot.
-- CSS/SVG/GSAP device used for motion structure.
-- Transition out.
-- Audio/rhythm note.
-- Quality risk.
-
-The hero frame is the timestamp where the scene best communicates its idea. If that frame is weak as a still image, the scene is not ready for animation.
-
-## 6. Visual Asset Planning
-
-If the metaphor needs generated bitmap source material, plan images before implementation. Count distinct visual worlds and independently movable foreground objects before choosing how many assets or sheets to generate. Do not default to one reusable background or to a duration-based asset count.
-
-- Background stage: atmosphere, depth, light, metaphor, or product context.
-- Center symbol.
-- Background texture or atmosphere.
-- Optional person/object anchor.
-- Support assets: symbol, texture, anchor, transition plate, semantic glyph, product/UI fragment, or motion accent.
-- Default vertical key visual for 9:16 work.
-- Horizontal key visual only when the brief documents a 16:9 target.
-
-Generated images should be sparse. Avoid baked-in text unless exact text is required. Compose final titles, subtitles, proof notes, masks, crops, parallax, focus pulls, and timing in HyperFrames.
-
-Each planned/generated image must specify:
-
-- Role: stage, symbol, texture, anchor, or transition plate.
-- Source: supplied, official, generated, or fallback pure-code.
-- Aspect ratio and target size.
-- Target scene/beat.
-- Layout contract and image ratio.
-- Focal subject location.
-- Quiet text zone.
-- Crop-safe subject zone, text rectangle, and subject rectangle.
-- Safe bottom/right/top platform boundaries for any overlaid text.
-- Local output path.
-- Forbidden content: no baked-in text, fake logos, generic icons, labels, diagrams, decorative clutter, or watermark unless required.
-- For support assets: shared style lock, isolation/cutout plan, safe zones, motion purpose, and deletion trigger.
-- For asset sets: visual-world inventory, component inventory, count reasons, sheet strategy, separate hero-object decision, manifest path, and dark/light cutout proof.
-
-Use Codex Image Gen by default after user confirmation when a generated bitmap is needed. Prefer HyperFrames/CSS/SVG for simple lines, marks, masks, traces, glyphs, and geometric accents. Inspect returned images against the layout contract before implementation. Recrop or regenerate if the quiet zone is busy, the subject lands in an unsafe crop area, or fake UI/text/logos appear. Preserve component sheets, crop every accepted item into an independent transparent asset, verify it on dark and light proof sheets, and run `validate_image_assets.mjs` before animation. Reference project-bound generated images only after they have been moved into the project asset folders.
-
-## 7. Optional Beat Map
-
-Use `BEAT_MAP.json` when timing matters.
-
-Include:
-
-- FPS.
-- Duration.
-- Beat/hit timestamps.
-- What should happen on each hit.
-- Optional narration word timings.
-
-Do not pretend automatic audio analysis has happened unless it actually has. Manual hit maps are acceptable for MVP work.
-
-## 8. Static Build
-
-Build static scene layouts before animation.
-
-Check:
-
-- Copy is legible.
-- Chinese copy uses deliberate line breaks, large type, and low character count per beat.
-- Visual hierarchy is obvious.
-- Background image has one clear role and is not generic wallpaper.
-- Support assets have roles, shared style lock, safe zones, motion purpose, and deletion triggers.
-- No support asset competes with the dominant text, symbol, product, or CTA.
-- Text is placed on a planned quiet zone, not over high-frequency details.
-- The selected layout contract matches the message and image shape.
-- The image ratio is standard and matches the intended placement.
-- The text rectangle groups title, support, CTA, subtitles, and proof notes into a readable hierarchy.
-- The subject rectangle keeps faces, product edges, UI detail, and symbols out of unsafe crop areas.
-- Safe bottom/right/top boundaries keep CTA, subtitles, support lines, and proof notes clear of vertical platform overlays.
-- Local contrast treatment is present where text crosses imagery.
-- Safe margins hold.
-- Vertical platform overlay zones are respected, especially bottom UI / CTA collision risk.
-- Text has max width, max lines, explicit overflow behavior, and stable breakpoints.
-- Long words, mixed Chinese/English copy, CTA labels, and subtitles stay inside their containers.
-- The composition does not depend on motion to make sense.
-- Motion paths preserve the quiet zone and keep readable copy away from unsafe mobile UI zones.
-- Transitions use the image subject, text axis, mask, light direction, or camera move as a bridge; they are not separate decorative effects.
-- Product or brand appears with enough weight.
-- The metaphor can be understood without explanatory icon labels.
-- The frame obeys the house style: black, sparse, cinematic, white/gray/warm gold.
-- No ordinary illustration, ecommerce banner, icon pile, generic neon tech, multicolor palette, or busy collage appears.
-
-## 9. Motion Build
-
-Add animation only after static layouts are strong.
-
-Motion should:
-
-- Order attention.
-- Reveal the metaphor.
-- Clarify cause and effect.
-- Reinforce audio hits.
-- Transition between ideas.
-- Avoid decorative noise.
-- Make the work impossible to reduce to static screenshots without losing meaning.
-
-For kinetic text relay promos, at least half of the scene changes should use directional or structural transitions: push left/right, push up/down, crop wipe, scan pass, split, compression, expansion, type-on handoff, or object-led cut. Fade may support a transition, but should not be the main transition language.
-
-Motion budget:
-
-- One primary motion idea per scene.
-- One optional secondary motion for atmosphere or focus.
-- Background motion should be slow push, parallax, focus pull, or light sweep only.
-- Main text should settle before it needs to be read.
-- Important reveals need a still hold, not continuous drift.
-- Scene transitions should overlap or share an anchor. Empty black is reserved for deliberate silence or final release.
-
-Motion craft requirements:
-
-- Build a labeled GSAP master timeline for multi-scene work.
-- Use CSS3/SVG devices for structural reveals when appropriate: masks, clip paths, scan paths, underlines, frames, connectors, or perspective layers.
-- Give important text a transition pattern beyond opacity and y-position.
-- Let hook or amplified keyword text occupy center or upper-center impact when it is the main subject.
-- Reserve lower-safe placement for CTA, proof notes, or subject-dominant frames.
-- Include at least one signature motion moment that belongs to the metaphor.
-- Score text/icon transition promos with the kinetic relay scorecard from `references/motion-craft.md`; target 100, below 90 blocks final delivery, and below 70 requires revision before render.
-
-## 10. Validate
-
-Run HyperFrames checks or the closest available substitute:
-
-```bash
-npx hyperframes doctor
-npx hyperframes lint
-npx hyperframes validate
-npx hyperframes inspect
-npx hyperframes snapshot <composition> --at <times>
-```
-
-For dense videos, snapshot hero frames and transition frames. A video can render while still failing visually; snapshots catch this earlier.
-
-## 11. Render And Review
-
-Render a draft before final:
-
-```bash
-npx hyperframes render --quality draft --output renders/draft.mp4
-npx hyperframes render --quality standard --output renders/review.mp4
-```
-
-Use Docker/high quality for final delivery if available and appropriate.
-
-Finish with `REVIEW_REPORT.md`:
-
-- What was produced.
-- Which checks passed.
-- Which checks could not run.
-- Watch notes by time range.
-- Remaining risks.
-- The next best edit.
-
-## Editing Protocol
-
-When revising, do not rewrite the whole composition unless the user asks for a new direction.
-
-Map feedback to the smallest change:
-
-- "Make it punchier" usually means timing, transition, or copy density.
-- "More cinematic" usually means less clutter, stronger type, longer holds, fewer effects.
-- "More clear" usually means script or hierarchy.
-- "The product gets lost" means layout, contrast, or product scale.
-- "Music doesn't hit" means beat map and timeline positions.
-
-After revising, rerun validation relevant to the changed layer.