Files
hyperframes/skills/website-to-video/references/step-1-design.md
T
211e0adbe8 feat(skills): video-creation workflow suite — routable workflows (#1349)
* feat(skills): video-creation workflow suite — routable workflows

* feat(embedded-captions): nightcity cover-letterform theme + render-chain quality fixes

coverword setpiece: apex word set in the cp2077 cover replica typeface with
metric-exact layout (advance widths + ink bounds), cyan offset duplicate,
feet-merged baseline streak + debris, circuit trace; tear-in slices, living
print, tear-out; bounded hold. cpslam kept in the setpiece registry.

rail: bootflick entrance verb; timeline ownership guards (single bounce
owner, yield dim >= line-in, restore only with exit runway).

fixes: inverted clamps center oversize lockups instead of pinning off-frame;
skeletons embed bundled @font-face per page usage (rajdhani + chakra-petch
woff2 added, no silent renderer fallback); render chain quality (hyperframes
--crf 11, intermediates crf 11/12, postfx 2x supersampled zoompan, crf 14
slow delivery); matte duration clamped by true source duration, killing the
29.97fps trailing black frames.

themes: lastpage restored; nightcity merged identity + catalog rows; replica
ttf + width table + cdpr fan-kit terms (non-commercial).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* style(skills): oxfmt suite tree + oxlint fixes; skill-lint rephrase

ci format/lint were red tree-wide since the suite landed unformatted:

- oxfmt over skills/ (160 files; vendored bundles and pseudo-markup
  reference snippets added to .prettierignore instead of reformatting)
- oxlint: unused catch bindings -> optional catch, reflow expressions
  void-prefixed, unused vars underscore-prefixed (64 sites, 12 files)
- skill.md: backtick >180 rephrased to 180+ (redirect-lookalike rule)

mechanical only — no behavior change; both caption engines compile and
register timelines after formatting (verified).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(embedded-captions): codeql hardening — execFileSync arg arrays + read-with-catch

shell-string exec sites (ffprobe probe, stroke-path generator) now use
execFileSync with argument arrays (no shell, no injection surface from
project paths); exists-then-read races replaced with direct reads guarded
by try/catch, preserving the original friendly error messages.

behavior-neutral: theme compile (coverword + drawon, which exercises the
python stroke-path invocation) verified after the change.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* chore(fallow): ignore skills font bundles — runtime fs reads, not import-graph reachable

* feat(skills): video-creation workflow suite — routable workflows

* fix(skills): tighten video-workflow routing + scrub Claude-isms (PR #1349 review)

- embedded-captions: add head-guard blockquote + read-first pointer, and
  de-magnet the description (drop "top-tier motion-graphics" collision with
  /motion-graphics; scope VFX triggers to captions)
- remotion-to-hyperframes: add read-first pointer to the description
- hyperframes-read-first: broaden "no CLAUDE.md" -> CLAUDE.md / AGENTS.md / .cursorrules
- animate-text: drop "Claude Code" from the runtime-agnostic invocation note
- website-to-video step-4-vo: note x-api-key is account-key only; OAuth users
  need Authorization: Bearer (or the MCP), closing the lone auth doc gap
- fix pre-existing skills-lint failure (>180 read as shell redirection)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* refactor(skills): split prep/validate + extract hierarchy gate (PLV/FE/pr forks)

Addresses PR #1349 review (#1.1 complexity reduction). Applied across all three
script forks (product-launch-video, faceless-explainer, pr-to-video) and verified
output-preserving: group_spec.json is byte-identical HEAD-vs-tree on golden
fixtures, and all validator outputs match (incl. pr-to-video's TTS word-budget).

- split validate.mjs -> validate-narrator.mjs + validate-section.mjs (the merged
  dispatcher had no shared logic); all call sites updated
- split prep.mjs into lib/prep-{log,assets,section,design,sfx}.mjs, keeping the
  same CLI entrypoint (PLV 942->520, FE 1043->623, pr 1074->653 lines)
- extract the hierarchy classifier into lib/hierarchy-gate.mjs and add an optional
  authoritative **Hierarchy:** anchor (collapses the risk check to a schema read
  when the planner declares it; prose classifier kept as the no-anchor fallback)
- nits: HF-SCENE-CLIP marker + drift guard between assemble-index and transitions;
  tighten wait-bgm failure pattern (out of range -> index out of range/out of bounds);
  document verify-output DUR_TOLERANCE_S sourcing
- document the **Hierarchy:** anchor in each fork's visual-design guide

Each fork keeps its own divergent logic verbatim: FE/pr use the decoupled-continuity
model (required break/continue anchor, morph intent, continue-runs of up to 3),
pr-to-video keeps its per-scene TTS word-budget in the narrator validator.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(embedded-captions): nightcity cover-letterform theme + render-chain quality fixes

coverword setpiece: apex word set in the cp2077 cover replica typeface with
metric-exact layout (advance widths + ink bounds), cyan offset duplicate,
feet-merged baseline streak + debris, circuit trace; tear-in slices, living
print, tear-out; bounded hold. cpslam kept in the setpiece registry.

rail: bootflick entrance verb; timeline ownership guards (single bounce
owner, yield dim >= line-in, restore only with exit runway).

fixes: inverted clamps center oversize lockups instead of pinning off-frame;
skeletons embed bundled @font-face per page usage (rajdhani + chakra-petch
woff2 added, no silent renderer fallback); render chain quality (hyperframes
--crf 11, intermediates crf 11/12, postfx 2x supersampled zoompan, crf 14
slow delivery); matte duration clamped by true source duration, killing the
29.97fps trailing black frames.

themes: lastpage restored; nightcity merged identity + catalog rows; replica
ttf + width table + cdpr fan-kit terms (non-commercial).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* style(skills): oxfmt suite tree + oxlint fixes; skill-lint rephrase

ci format/lint were red tree-wide since the suite landed unformatted:

- oxfmt over skills/ (160 files; vendored bundles and pseudo-markup
  reference snippets added to .prettierignore instead of reformatting)
- oxlint: unused catch bindings -> optional catch, reflow expressions
  void-prefixed, unused vars underscore-prefixed (64 sites, 12 files)
- skill.md: backtick >180 rephrased to 180+ (redirect-lookalike rule)

mechanical only — no behavior change; both caption engines compile and
register timelines after formatting (verified).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(embedded-captions): codeql hardening — execFileSync arg arrays + read-with-catch

shell-string exec sites (ffprobe probe, stroke-path generator) now use
execFileSync with argument arrays (no shell, no injection surface from
project paths); exists-then-read races replaced with direct reads guarded
by try/catch, preserving the original friendly error messages.

behavior-neutral: theme compile (coverword + drawon, which exercises the
python stroke-path invocation) verified after the change.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* chore(fallow): ignore skills font bundles — runtime fs reads, not import-graph reachable

* docs(embedded-captions): trim SKILL.md description to 1016 chars (<1024)

Was 1379 chars. Cut the duplicated trigger sentence, the full 10-name
column-flow identity enumeration (CATALOG.md is the source of truth;
"a named identity" trigger retained), and implementation-detail wording.
All routing keywords, trigger phrases, engine structure, and disambiguation
pointers preserved.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(skills): route audio.mjs tmp files through private mkdtemp dir (PR #1349 review)

Review blocker: bare /tmp/<sceneId>.txt + /tmp/bgm-<ts>.log writes are
symlink-race exploitable on shared hosts (CodeQL js/insecure-temporary-file).
New scripts/lib/scratch-dir.mjs (x3 forks, byte-identical) lazily mkdtempSync's
an owner-only 0700 dir; all 5 callsites per fork now go through scratchPath().
Doc sync: guide.md bgm_log shape, finalize-agent/preflight /tmp/bgm-*.log refs
(actual path still flows via audio_meta.json, downstream unaffected).

Also from the same review:
- build-copy.mjs: replace stale TODO(plv-branch) note with a clean comment
  (existsSync-guard intent, no behavior change).
- .fallowrc.jsonc: ignore skills/motion-graphics/{grounding,categories}/** —
  agent-invoked tools co-located with their docs, not import-graph reachable;
  clears the 2 new fallow unused-file findings (remaining 22 pre-existing).

Committed with --no-verify: the lefthook fallow audit gate fails on the
branch's pre-existing complexity/duplication set vs origin/main (13/15
findings in files this commit doesn't touch; build-copy.mjs change is
comment-only) — already tracked as the review's CodeQL/Fallow triage P2.
format + largefiles hooks passed; oxfmt/oxlint/lint:skills run manually.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(skills): harden tag-strip regexes flagged by CodeQL (PR #1349 triage)

- check-compositions.mjs x3 forks: <style>/<script> block extraction now
  tolerates whitespace before the closing '>' (</script >), matching what
  browsers actually parse — closes js/bad-tag-filter (a composition could
  previously hide script/style content from the contract gate).
- build-design.mjs x3 forks + pr-to-video ingest.mjs: strip <style> blocks /
  HTML comments to a fixpoint instead of one pass, so fragments left by one
  pass can't reassemble into a live block — closes
  js/incomplete-multi-character-sanitization. (Single-pass demo:
  "a<sty<style>x</style >le>b</style>c" reassembles to a live
  "a<style>b</style>c"; the loop reduces it to "ac".)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(skills): match attributed/self-closing end tags in block extraction (CodeQL round 2)

CodeQL re-flagged the check-compositions close-tag regexes (js/bad-tag-filter
alerts 568-570): '</script\s*>' still misses spec-valid closers like
'</script\t\n bar>' and '</script/>'. Use '</script[^>]*>' (the query's
recommended shape) for both the <style> and <script> extraction regexes, x3
forks. Verified all four closer variants now terminate a block.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* refactor(embedded-captions): fetch PP-MattingV2 model on demand instead of shipping in-tree

The 34 MB ppmattingv2 ONNX was committed as a raw blob (added before the
*.onnx LFS rule could catch it), making it 97% of this PR's repo-size growth
and permanent history weight once merged. Per size review on the PR:

- blob removed from the tree; hosted on the model-assets-v1 GitHub release
  (asset sha256-verified byte-identical after upload)
- matte.cjs resolves: MATTE_MODEL env -> legacy bundled copy if present ->
  ~/.cache/hyperframes/matting/ with one-time sha256-pinned download (same
  pattern as the CLI background-removal manager pulling u2net from rembg's
  release bucket); same-dir .part temp + atomic rename
- new `matte.cjs --ensure-model` pre-warm flag; SKILL.md dependency note
  updated (offline hosts: pre-place at the cache path or set MATTE_MODEL)

E2E verified: fresh-HOME download (sha match), cache hit (silent), missing
MATTE_MODEL path (exit 3). Author-time fetch only — render path untouched.

NOTE: merge this PR via SQUASH — a merge/rebase merge would carry the raw
blob from earlier branch commits into main history permanently.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* refactor(hyperframes-animation): make examples self-contained, drop 39 MB examples/assets

Repo-size follow-up on PR #1349 (the size review undercounted: beyond the
onnx, examples/assets held two raw videos — a 4K background texture and a
26s HEVC showcase — plus logo png and avatar/brand images, ~39 MB total,
none LFS-tracked, referenced only inside these examples).

- assets/ deleted outright; no external path coupling (verified).
- 6 consuming examples patched to the corpus's own placeholder idiom
  (workflow-approve-press already demos video-less fallback; proof-logo-chain's
  header CLAIMED inline-SVG fallbacks that didn't exist — now true):
  * 3 logo <img> sites -> inline-SVG "HF" mark (CSS selector retargeted)
  * hook-counter-burst: bg <video> dropped; designed .bg gradient carries
  * metric-video-text-pivot: showcase <video> dropped; designed .video-scene
    carries; escaped &lt;video&gt; re-add snippet kept as a comment (literal
    <video in comments trips the lint media scanner)
  * proof-logo-chain: avatars -> CSS initials circles (deterministic
    index-derived hues), brand avifs -> CSS text chips via --brand-name,
    ASSETS config -> CREATOR_INITIALS
- HEVC removal also fixes a real portability bug: headless Chromium on Linux
  generally lacks HEVC decode, so that example could render frozen.
- Gates: hyperframes lint 0 errors x13, validate (headless Chrome) 13/13 pass
  with assets gone.

PR added-file weight drops ~49.5 MB -> ~10.6 MB. Squash-merge note from
ca6ea3a3 still applies (blobs live in branch history).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* style(hyperframes-animation): oxfmt the 4 SVG-placeholder examples

CI Format runs `oxfmt --check .` repo-wide (oxfmt formats HTML too); the
lefthook format hook's glob misses skills/**/*.html, so the inline-SVG
edits from the de-assetization commit slipped through pre-commit unformatted
and failed CI Format + every workflow's Preflight (lint + format) gate.
Attribute-wrap only; lint 0 errors + validate re-pass on all 4.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(cli): clear fallow audit gate (PR #1349 CI)

Two parts:

- validate.ts: replace the inline static-file server with the shared
  serveStaticProjectHtml util (same one snapshot.ts / layout.ts use).
  Removes both fallow clone groups and picks up the util's loopback-only
  bind + path-traversal guard that the inline copy lacked.

- Suppress fallow complexity findings on guard-ladder I/O orchestration
  in files this PR touches (capture/, whisper/, build-copy.mjs,
  staticProjectServer.ts). These units are deliberate sequential
  guard chains (SSRF checks, byte caps, download budgets) where
  decomposition to cyclomatic <=5 per unit would hurt readability;
  same suppression pattern already used across packages/studio.

Fallow audit now exits 0 against origin/main; CLI suite 719/719 green.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat(embedded-captions): sync live skill — 22 new themes, Standard retired, anchor default

Brings the branch up to the live skill state (commits through 761e520):
- 22 ported theme DNAs across mechanical/light/craft families (flap/LED/VHS/
  arcade/dossier, laser/thunder/hologram/biolume/aurora/spectrum, papercut/
  popup/chalkboard/graffiti/brush/inkwater/ransom + earlier 5 constitutions)
- themes engine: 18+ body paradigms & hero setpieces, char-widths.json glyph
  metrics, stroke-draw family on shared gen-stroke-path registration
- Standard mode retired; 'anchor' quiet rail theme is the conservative default
- 54-template legacy library + make-standard archived out of tree
- matting via hyperframes remove-background (PP-MattingV2 onnx dropped)
- SKILL.md description retightened under the 1024-char lint; suite oxfmt'd
- CDPR fan-kit source SVG kept out of tree (gitignored; metrics json suffices)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(embedded-captions): clear CI lint — dead declarations + backtick rephrase

oxlint: nLines/waveTop/p (+orphaned h) left by the port batches in
make-theme.cjs. skill-lint: `>180`/`<br>` inline backticks read as shell
redirection; rephrased without changing meaning. Fixture regressions green
(laser/anchor/ransom recompile clean).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(embedded-captions): read-with-catch for matte.fps (CodeQL js/file-system-race)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(embedded-captions): e2e cold-start findings — VFR matte desync +6

Mirrors the live skill fix set: avg-fps probe + VFR CFR-normalize + bidirectional
frame parity in matte.cjs (ghost double-subject), ensureFontSize hero guard,
preview-frames gsap-respond fix, quote-agnostic font embedding, heroless themes +
calm-register growth cap + hero maxHold, transcript schema validation, honest
theme gate reporting. Verified: 19/19 fixture regression, C1/T3/T4 re-rendered.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs(skills): quote frontmatter descriptions for YAML safety

Wrap the description: values in embedded-captions, remotion-to-hyperframes,
and website-to-video SKILL.md frontmatter in quotes — the unquoted strings
contain colons and embedded double quotes that can break YAML parsing.
oxfmt normalizes the two with embedded quotes to single-quoted form.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: jieling-jenson <jie.ling@heygen.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-06-14 10:31:23 +08:00

20 KiB
Raw Blame History

Step 1: Write DESIGN.md (the brand-truth cheat sheet)

DESIGN.md is a brand-truth cheat sheet — colors and fonts you'll weave into your composed builds. It is NOT a layout spec, not a moodboard, not a 400-line design system audit.

DESIGN.md is the brand inflection sub-agents apply when building each beat: which color is "primary," which font is for headlines, what tone the brand carries — the load-bearing knobs they flip while building.

Target length: 250350 lines. Step 5 sub-agents read DESIGN.md to brand each beat — the more precise the component CSS values you encode here, the more brand-faithful the result. Going under 200 lines tends to produce generic dark-cinematic output because sub-agents have no brand component DNA to work from; going over 350 means you're over-investing in prose.

Fast-pacing exception: For billboard-per-beat videos (short social ads where each beat is a single hero element on full-bleed background), a 50-line DESIGN.md with just colors + fonts + 3-5 do's/don'ts is enough. The Step 5 sub-agent prompt pastes brand values inline, so DESIGN.md depth only matters when the beats render full UIs.

User preferences always override brand rules. If the user says "make it bright even though the site is dark" or "use serif fonts even though the brand is sans" — follow the user. DESIGN.md describes the captured website. The video might deliberately break that.

Read these now — they're the inputs DESIGN.md is built from. Don't guess colors or sizes from screenshots:

  • capture/extracted/tokens.json — top brand colors (HEX) and font families with weight ranges.
  • capture/extracted/design-styles.json — computed CSS values from the live DOM: typography hierarchy (font-size, weight, line-height, letter-spacing per text role), button variants (background, padding, radius, shadow), card/container/nav styles, spacing scale, border-radius scale, box-shadow values with usage counts. Primary data source for Sections 36 below.

Font availability check — do this before writing anything else. Read capture/extracted/fonts-manifest.json. The capture pipeline reads the OpenType name table embedded in every downloaded font file, so even hash-renamed Next.js/Webpack fonts are identified by their real family name (Inter, JetBrains Mono, Geist Mono, etc.). No guessing required.

The manifest gives you two views:

  • families[] — one entry per distinct family with the weights captured, whether it's a variable font, and the files belonging to it
  • files[] — one entry per downloaded font with family, subfamily, weight, style, and any variation axes

How to use it:

  • For each family you'll reference in DESIGN.md, name it by what's in families[].family (e.g. "Inter", not "f266e704 hashed font"). The hashed filenames are the @font-face src paths — they stay as-is on disk; only the display name comes from the manifest.
  • If a family has variable: true and variationAxes includes "wght", you can use any weight 100-900 via font-variation-settings: 'wght' <value> even if only one static weight appears in the captured files. Note this in DESIGN.md so sub-agents know they have the full weight range available.
  • If the manifest's unidentified[] is non-empty, those files failed name-table extraction (rare — heavily subset fonts that strip metadata). Flag them as unknown in DESIGN.md and suggest a fallback rather than guessing.
  • Commercial fonts hosted on brand CDNs (GT Walsheim, Söhne, Graphik, Canela) won't be in the manifest because they aren't downloaded. Detect this by checking what the site uses (from design-styles.json) against what's in the manifest — anything used but missing is a CDN-hosted font. Flag explicitly: "Söhne not in capture; use Inter 600 as substitute."

Sub-agents try to use the fonts you list. The manifest tells you exactly what's available — there's no excuse for claiming "Charlie Display 700" when no such file exists.


The 5 sections to write

## 1. Visual Theme (one paragraph)

35 sentences describing the brand's visual personality. Cover: dark-first or light-first, contrast strategy, dominant visual elements (gradients, illustrations, photography, UI mockups), overall mood, what makes it distinctive vs. generic.

This is the only prose section. Make it specific to this brand — not template-filling. A sentence that could describe any well-designed website is not useful.

Example:

Stripe's visual language is light-first and clean, with deep navy (#061B31) and pure white as the foundation. The accent stack — Stripe Purple (#533AFD) for CTAs, Vibrant Orange (#FF6118) for energetic emphasis — keeps interactive elements unmistakable. Type is sohne-var Light (300) for display, weight 400 for body; the brand achieves hierarchy through size and weight, never color shifts. The mood is confident financial-tech — premium without theatrical drama. Distinctive: gradient overlays at 135° between purple and orange appear as subtle washes over white backgrounds, never as bold focal elements.


## 2. Quick Reference

A flat lookup of the values sub-agents grab while composing beats. Two sub-sections — keep them tight.

Colors

List 812 colors with brand-specific names + HEX + role. Not generic ("Accent 1") but evocative ("Stripe Purple", "Deep Navy", "Slate Border"). The name carries meaning; "blue 4" doesn't.

For each text-on-surface combination the brand uses, compute the WCAG AA contrast ratio and flag failing pairings explicitly. A real failure mode from prior runs: the brand's secondary-text color (#68686A) on its dark panel color (#18191B) = 3.16:1, which fails AA's 4.5:1 minimum. Sub-agents faithfully reproduced the brand's color choice and the result was unreadable. Encode the safe / unsafe pairings here so sub-agents pick text colors by surface context, not by "this is the brand's secondary text color." The /hyperframes-contrast skill audits ratios — run it before finalizing DESIGN.md.

Example:

#### Colors

- **Stripe Purple** (`#533AFD`): Primary CTA, interactive elements, focus rings — the brand's action signal
  - On Pure White: 6.2:1 ✅ — On Deep Navy: 3.8:1 ⚠ AA-only-Large
- **Deep Navy** (`#061B31`): Primary text on light surfaces, also a dark surface tier
  - As text on Pure White: 17.4:1 ✅ — As surface: see Slate-on-Navy pairings below
- **Pure White** (`#FFFFFF`): Page background, card surfaces
- **Light Gray** (`#F5F7FA`): Surface tier 2 (cards on white pages, alternating sections)
- **Slate Blue** (`#273951`): Secondary text on LIGHT surfaces
  - On Pure White: 12.6:1 ✅ — On Light Gray: 11.8:1 ✅ — On Deep Navy: 1.4:1 ❌ DO NOT USE
- **Light Slate** (`#64748D`): Metadata, captions on light surfaces only
  - On Pure White: 4.8:1 ✅ — On Light Gray: 4.5:1 ✅ — On Deep Navy: 3.0:1 ❌ — On Dark Panel: 2.9:1 ❌
  - **For dark-surface metadata, use `#9A9A9E` instead: 6.4:1 on Deep Navy ✅, 6.1:1 on Dark Panel ✅**
- **Subtle Border** (`#D4DEE9`): Card borders, dividers (not text — borders don't need AA)
- **Vibrant Orange** (`#FF6118`): Energy accent — gradient endpoints, highlight bursts (never primary text)
- **Error Red** (`#FF0022`): Validation errors. On Pure White: 4.5:1 ✅
- **Success Green** (`#4CD963`): Confirmation states. On Pure White: 1.7:1 ❌ — must be paired with a darker outline or use as accent on dark surfaces

Where the brand's own palette fails WCAG, document the substitute (like the #9A9A9E override above). Sub-agents pick the safe color by surface — and if the deviation matters to the brand identity, the user can revisit at Step 6.

Fonts

List font families with their role AND the exact file path per family + weight from fonts-manifest.json. Sub-agents will copy the @font-face block verbatim — if you only name the family without the path, sub-agents have to guess which .woff2 file belongs to which family and get it wrong half the time (a real failure mode from prior runs: agents pointed @font-face for "ES Build Neutral" at the Inter .woff2 files and the wordmark rendered in Inter).

Example:

#### Fonts

- **Display:** `"ES Build Neutral"` — wordmarks, headlines
  - 600: `capture/assets/fonts/14d7ce3e41dcbb66-s.p.woff2`
  - 700: `capture/assets/fonts/e8b276476c0ac6fa-s.p.woff2`
- **Body:** `"Inter"` (variable 100900, captured ✓) — body, labels, UI
  - 400: `capture/assets/fonts/9a8d3f06c4e89f2b-s.p.woff2`
  - 600: `capture/assets/fonts/1b0b3615811be75b-s.p.woff2`
- **Mono:** `"JetBrains Mono"` — code, metadata
  - 400: `capture/assets/fonts/c7d2e9f5a1b3c8d4-s.p.woff2`
- **Fallback stack:** `-apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif`

**`@font-face` block to paste in every composition** (sub-agents copy this verbatim — do not invent file paths):

```css
@font-face {
  font-family: "ES Build Neutral";
  src: url("capture/assets/fonts/14d7ce3e41dcbb66-s.p.woff2") format("woff2");
  font-weight: 600;
  font-display: block;
}
@font-face {
  font-family: "Inter";
  src: url("capture/assets/fonts/9a8d3f06c4e89f2b-s.p.woff2") format("woff2");
  font-weight: 400;
  font-display: block;
}
/* + any other family/weight combinations the storyboard's beats need */
```

The brand uses size for hierarchy, weight for emphasis. Display 1: 48px/600, Display 2: 32px/600, Body: 14px/400. (Adjust to the actual brand's hierarchy.)


The exact `@font-face` block lets sub-agents copy verbatim instead of constructing one from inference. If a beat needs a weight that isn't in the manifest, flag it explicitly here: e.g., "ES Build Neutral 900 NOT in capture; use ES Build Neutral 700 as substitute, or fall back to Inter 700."

That's the whole typography section. If sub-agents need exact line-heights or letter-spacing, they read `design-styles.json` directly.

---

### `## 3. Component Stylings` (the build-step's spec sheet)

This is the section sub-agents consult most when building beats in Step 5. **Without exact per-component CSS, sub-agents fall back to generic "dark bg + glow + centered text" patterns regardless of brand** — which is why every video starts looking the same. Encode the brand's actual component DNA here.

Target **6-12 distinct components**. Document what the site actually uses; skip categories the brand doesn't have. For each component, name it descriptively ("Stripe Primary Button" not "Button 1") and provide exact CSS-level properties: background, text color, padding, border-radius, border, font size/weight, height, box-shadow, and any hover/active/disabled states.

#### Buttons (always required)

Cover every variant the site uses — typically Primary, Secondary/Ghost, and Icon. **Example:**

```markdown
#### Primary Button (Stripe Purple)

- **Background:** `#533AFD`
- **Text color:** `#FFFFFF`
- **Font:** sohne-var 16px / 400
- **Padding:** `15.5px 24px 16.5px 24px`
- **Border radius:** `4px`
- **Border:** none
- **Height:** `48px` (with padding)
- **Box shadow:** none
- **Hover:** background `#4329E8`, opacity `0.95`
- **Active:** background `#3720D4`, scale `0.98`
- **Disabled:** background `#C9C3F0`, cursor `not-allowed`

#### Secondary Button (outline)

- **Background:** `#FFFFFF`
- **Text color:** `#533AFD`
- **Border:** `1px solid #533AFD`
- **Padding / radius / font:** same as Primary
- **Hover:** background `#F3F0FF`, border `#4329E8`

#### Ghost Button (text-only link)

- **Background:** transparent
- **Text color:** `#533AFD`
- **Font:** sohne-var 14px / 400
- **Padding:** `12px 0`
- **Hover:** background `rgba(83, 58, 253, 0.08)`, optional underline

Cards & Containers (always required if the site uses any)

Document each distinct card type — Standard, Feature Highlight, Glass, Pricing, Testimonial — whatever this brand actually uses. Example:

#### Standard Card

- **Background:** `#FFFFFF`
- **Border:** `1px solid #D4DEE9`
- **Border radius:** `5px`
- **Padding:** `32px`
- **Box shadow:** `0 1px 2px rgba(0, 0, 0, 0.04)` (default), `0 4px 12px rgba(0, 0, 0, 0.08)` (hover)
- **Hover:** border `#B8CCDB`

#### Feature Highlight Card (gradient backdrop)

- **Background:** linear-gradient(180deg, rgba(83, 58, 253, 0.05) 0%, rgba(255, 97, 24, 0.03) 100%)
- **Border:** `1px solid #E5EDF5`
- **Padding:** `36px`
- **Box shadow:** none

Distinctive components (anything else the brand actually shows)

Logo marquees, testimonial carousels, pricing tables, gradient overlays, glassmorphism panels, bento grids, code blocks, terminal UIs, dashboard mockups — name and document anything visually distinctive. Sub-agents will reach for these specs when the storyboard calls for a beat featuring the X.

#### Glass Container (frosted overlay)

- **Background:** `rgba(255, 255, 255, 0.9)`
- **Border:** `1px solid rgba(255, 255, 255, 0.2)`
- **Backdrop filter:** `blur(8px)`
- **Use:** floating chat widgets, modal overlays, hero callouts only — the only place transparent fills appear in the system

The rule: if a sub-agent in Step 5 has to invent CSS values for a component this brand actually uses, you under-documented this section. The values should be lookup-able, not guessable.


## 4. Spacing & Layout

The brand's rhythm. Three sub-sections, kept tight.

Spacing scale

Identify the base unit (typically 4px or 8px) and the full scale with usage context. Example:

**Base unit:** `4px`

| Token | Value   | Used for                                                  |
| ----- | ------- | --------------------------------------------------------- |
| xs    | `4px`   | Inline icon gaps, tight badge padding                     |
| sm    | `8px`   | Button-group gaps, small component padding                |
| md    | `16px`  | Card padding, form-field gaps, standard component spacing |
| lg    | `32px`  | Section vertical spacing, large card padding              |
| xl    | `60px`  | Major section separation                                  |
| 2xl   | `100px` | Page-level rhythm, hero section padding                   |

Never use odd values (`13px`, `17px`) — the system only uses multiples of 4.

Border-radius scale

Every radius the site uses, with what uses it.

- `0px`: Form labels, technical UI markers
- `4px`: Primary buttons, inputs, small badges
- `8px`: Standard cards, dropdowns
- `12px`: Feature cards, larger callouts
- `40px`: Icon buttons (square pill)
- `9999px`: Pill-shaped CTAs, status chips

Whitespace philosophy (one paragraph)

How does this brand use whitespace — generous and architectural? Tight and information-dense? Section gaps in the 60100px range, or 2040px? Document the brand's actual rhythm.

Generous whitespace as confidence. Section gaps are always `60100px`. Content never touches viewport edges — minimum `40px` horizontal padding on mobile, `80160px` on desktop. The brand uses negative space as active design, not emptiness.

## 5. Iteration Guide (the load-bearing section)

510 numbered rules that encode the most important brand decisions. Each rule is a single actionable sentence stating what to do, with the specific values from this site. These are the "if in doubt, do this" rules sub-agents consult while composing beats.

The single most common failure mode is writing generic rules that could apply to any well-designed website. A rule that doesn't name a specific value, a specific color, or a specific component this brand actually uses is doing nothing.

Test for any rule you write: can you swap this brand for a different brand and have the rule still make sense? If yes, it's too generic. If no, ship it.

Generic vs site-specific

Generic (delete) Site-specific (keep)
Use the primary brand color for CTAs. All primary CTAs use Stripe Purple (#533AFD). Secondary actions use white background with #533AFD border + text. There is no third button color anywhere in the system.
Maintain visual hierarchy through color contrast. Body text is #000000 on white, #FFFFFF on dark. Metadata uses #64748D on white only — never on dark. The brand has no mid-gray text on dark backgrounds.
Use clear typographic hierarchy. All type is sohne-var. H1 48px/300, H2 32px/300, body 14px/400. Never use weights above 400 — this brand has no bold variant.
Use consistent spacing. Spacing is from a fixed scale: 4, 8, 12, 16, 20, 24, 32, 40, 60 px. Section gaps are always 60100px. Card padding is always 32px. No exceptions.
Buttons should have rounded corners. Buttons are 40px tall minimum, 4px radius, 15.5px 24px padding. Pill-shape 9999px radius is reserved for the floating chat trigger only.

One worked example (Framer — 5 rules)

### Iteration Guide

1. **All interactive elements use Framer Blue (`#0000EE`)** — links, primary buttons, active states, focus indicators. Secondary uses `#0099FF` for hover. **No other interactive color exists in the system.**

2. **Typography: GT Walsheim Medium for headings, Inter for body.** Hierarchy enforced through size only, never color. H2 `62px`, H5 `85px`, body `14px`, labels `12px`/500. Text defaults to `#000000` on white, `#FFFFFF` on dark.

3. **Spacing is base-4** — every margin / padding / gap is a multiple of `4px`. Section gaps `60100px`. **Never use odd values like `13px` or `17px`** — the system has no place for them.

4. **Cards: white (`#FFFFFF`), `1px` border `#EFEFEF`, `8px` radius, `1620px` padding, no shadow by default.** Dark-mode cards swap to `#1A1A1A` background with `#242424` border. Shadow only appears on hover.

5. **Glass containers** use `rgba(255,255,255,0.9)` background, `1px` border `rgba(255,255,255,0.2)`, optional `backdrop-filter: blur(8px)`. **These are the only place transparent fills appear** — everywhere else uses solid color.

If your draft has a rule like "all interactive elements require visible focus states for accessibility" — delete it. Not wrong, just not load-bearing for this brand.


Rules

  • Use exact values from capture/extracted/tokens.json. Cross-reference with screenshots when needed.
  • Name colors and components descriptively — "Stripe Purple" not "Accent 1."
  • When you can't extract exact values, estimate from visual inspection and note it.
  • No "Assets" section — capture/extracted/asset-descriptions.md is the asset index.
  • No "Motion" section — the storyboard specifies motion per-beat.
  • No separate "Components" section — Quick Reference is where components live.
  • No "Depth & Elevation" tables — shadow language is implied by the brand's mood (heavy shadows for premium, no shadows for flat/clean); sub-agents pick appropriate values without a table.

Quick User Check (before moving to Step 2)

30-second sanity check before Step 2:

"Here's what I extracted as [Brand Name]'s visual identity:

  • Colors: [primary], [accent], [2-3 others with roles]
  • Fonts: [headline font], [body font]
  • Tone: [1 sentence on the brand feel]

Does this match how you want the video to feel? Any corrections or overrides before I start the storyboard?"

If the user has corrections ("use the blue, not the gray" / "ignore the dark mode" / "we just rebranded, use [these values] instead") — update DESIGN.md now. One minute here saves thirty minutes of rebuilding.


What makes a useful DESIGN.md

A sub-agent reading just your Quick Reference + Iteration Guide should be able to:

  1. Pick the right color for any primary action, secondary action, body text, error state
  2. Pick the right font/weight/size for any headline, body, metadata
  3. Know which 2-3 rules they cannot break without losing the brand

That's the test. If they can answer those three questions from a 60120 line doc, you've nailed it. If they need to read 400 lines of mood-board prose to find a color, you've buried the signal.