OKLCH scale engine
Eleven-stop ramps with brand-tinted neutrals. Pin the seed to any step; edit stops in HSL or OKLCH without breaking monotonic lightness.
AgentStylusPalette
A deterministic brand-color stack for the agent era
OKLCH scales, harmony theory, and contrast proof — expressed as a stable API agents can call and designers can refine.
Field note · 01
Color keeps getting reinvented in the wrong room.
Agents invent. Designers refine. Products ship. Someone has to own the system between them.
Palette turns one seed into scales, roles, harmony, audits, and tokens — so the handoff isn’t a hex in a chat.
agent“Brand color is
#FF7A3D. Done.”
design“That’s a swatch. Where’s 50–950? Neutrals? Status? Can body text pass?”
ship“We exported three CSS files. None of them match Figma. The CTA fails AA.”
Capabilities
What the stack actually computes
Not a mood board. A reproducible pipeline from brief → brand system → audit → export.
Role-native ramps
Secondary, tertiary, and status (success / warning / danger / info) as first-class scales — parallel to primary, not bolted on.
Harmony ranking
Theory strategies scored by perceptual separation and audit fitness. Industry weights live in catalogue config, not hard-coded switches.
Contrast science
WCAG AA/AAA and APCA |Lc| targets, pair findings with auto-fixes, and an N×N matrix across brand, neutrals, and theme surfaces.
Mode physics
Solid, neon (OKLCH box + CTA APCA), and gradient with scrim sampling — each with explicit pass/fail gates.
Live evaluation
Full-bleed mockups and gallery views act as the visual unit test for every token change.
Dual-axis export
Package × notation: CSS, Tailwind, tokens, agent JSON — Hex, RGB, HSL, OKLCH — named from your document title.
Grounding inputs
Industry presets, website scrape (when CORS allows), and image extraction — start from seed, recipe, or evidence.
Interface contract
Built so agents can cite their work
Pure ESM library — no DOM. Same brief, same output. Provenance lists algorithms and parameters for audit trails.
Input
BrandBrief → seed | presetId | extracted[]
Output
BrandSystem → scales · harmony · audit · tokens
- suggestBrand / suggestCombinations — ranked variants with separation & score
- auditPalette / buildContrastMatrix — WCAG or APCA findings
- exportDesignTokens — agent.json and design packages
- OASF record — discovery envelope for skill/domain registries
Start
Run the pipeline in your browser
Seed → system → checks → export. No account wall between you and the model.