Design skills

Build AI features and apps on the major model providers (OpenAI, Anthropic/Claude, and Google Gemini) the model-agnostic way, so you can swap providers without rewriting.

Get the real images, icons, illustrations, fonts and footage a build needs, decide source versus generate versus commission, pull from licensed libraries, verify every licence and attribution requirement, treat assets into the palette, and record the provenance so it ships with the work. Owns sourcing and provenance; for art direction of generated imagery use as-design-image-styles.

Move creative assets between studios while preserving provenance, transparency and editability. Not a standalone image prompt or generic workflow scheduler.

Your on-demand branding consultant: get expert, reasoned answers to any brand DECISION instead of a one-shot kit.

Populate and verify a creative platform brand from an existing brand kit. Discovery alone stays with as-core-brand-discover; creating a new identity stays with as-design-brand-starter.

From a brand name + 3-5 vibe words, produce a complete mini brand book (positioning one-liner, color palette + type pairing, logo/mark concepts, and a voice mini-guide) assembled into one client-ready package (a $1-5k starter deliverable).

Work a single component to a finished state, generate real variants, prototype the interaction, then deliberately break it against long content, empty data, slow networks, errors, narrow widths, keyboard-only use and reduced motion. Returns the component plus the evidence it survives.

Produce concept and editorial illustration for a build, decide illustration versus photograph versus diagram, hold one visual system across a whole set, recolour and adapt licensed illustration into the palette, and keep a reusable library with its provenance. Owns illustration as a system; for photographic generation use as-design-image-styles, for explanatory schematics use as-design-diagram.

Give structured, severity-tagged design feedback on a screen, mockup, or built component, usability, hierarchy, consistency (with the a11y lens routed to the WCAG gate). Critiques against **the sender's** design system (this repo's tokens when the artifact is this app).

Build a reusable visual theme for the things you hand over (decks, reports, proposals, one-pagers, posters, certificates, covers) then compose and render pieces from it through whichever design tool is actually connected (Canva, Adobe Express, Gamma, an HTML pipeline, or a document toolchain).

Create an editorial-grade diagram (architecture, flowchart, sequence, state machine, ER/data model, swimlane, org chart, timeline, Sankey, Wardley map, user journey, deployment, dependency graph, UML class, database schema, and 24 more types) as a self-contained branded HTML/SVG figure, or redraw an existing draw.io/Mermaid diagram into presentable form.

Translate Figma designs into production-ready front-end code with high visual fidelity, pull design context, screenshots, variables/tokens, and assets via the Figma MCP, map them to the app's real design system (Tailwind/shadcn/tokens, not hardcoded values), and build accessible, responsive components that match the design 1:1.

Build or revise a coordinated founder-led visual campaign from consented reference photos and brand assets, including YouTube thumbnails, offer ads, social/profile covers, and website headers with editable sources, approval continuity, deterministic exports, QA evidence, and a resumable handoff.

Generate a developer-handoff spec from a design (layout, exact tokens, component props/API, interaction states, responsive breakpoints, edge cases, and motion) so engineering builds it once, correctly. Produces the buildable spec; for the WCAG gate use as-core-app-responsive-a11y, for building it use as-ops-premium-ui-builder.

Identity-preserving, visual-first portrait + photo-presentation studio, professional/personal-brand headshots (LinkedIn, founder, speaker, author, press, bio), tasteful social/dating profile photos, photo cleanup, and infographic presentation reports (face, hairstyle, outfit/color, grooming, profile-photo) with improved same-person previews and presentation-only scores.

Design category hub: routes to the right leaf for logo, brand, merch, diagram, visual-modes, visual-prompt-library, photoshoot, QR, theme, critique. Entry point for the design domain.

Apply print and retro treatments to an image you already have, 16-bit sprite, fat pixel, bootleg print, ASCII, green terminal, risograph, 1-bit Atkinson dither, punk collage, CMYK rosette dots, halftone print. Deterministic pixel math on our own engine, so composition is preserved by construction, byte-for-byte reproducible, $0 per image at any batch size.

Run a standing intake that turns what you see into what you can build, capture references daily with the read attached, file them by structural dimension rather than by vibe, and periodically build a small deliberate exercise against one captured idea so the input becomes capability. Pairs with as-design-reference-mining, which mines one reference for one build.

Run one cross-discipline review of a screen, flow, or interface change and return a single ranked verdict, accessibility, layout, hierarchy, density, affordance, typography, colour, polish and motion, each finding pinned to a location, cited to a named law, and priced by user impact.

Turn a system, process, Brain section, or pile of notes into a portable, diff-able visual graph (a valid JSON Canvas (.canvas) file (jsoncanvas.org 1.0 open spec) nodes[] + edges[]) laid out as a mind-map, flow, or architecture diagram, grounded in real source content.

Protect a surface an owner has already approved by eye. Refuses cross-cutting visual sweeps, converts them into a shown sample, and separates zero-movement refactors from ones that move pixels.

Logo and the full identity system, mark, wordmark, lockups, favicon, one-colour, reversed. Reads the live site first.

Author a whole expert method as a sellable Mega Prompt (role, anti-roles, phases, gates, rubric, refuse list) then hand it to as-design-megaprompt-compiler to encode as versioned JSON. One-prompt rewrites stay as-design-prompt-engineering; encoding an already-written method stays the compiler; chaining different skills stays as-core-workflow-builder.

Encode an already-authored expert method as a versioned JSON MegaPrompt spec, schema-validated, compiled per model (Claude first), inference-traced, rubric-scored. Ends prompt drift.

Design merch your audience will actually wear (T-shirts, hoodies, hats, stickers, mugs) that look professional, not amateur print-on-demand.

Run a real naming process for a company, product, offer, or feature, brief the claim the name must carry, generate a hundred candidates across every name type, screen on the six jobs (sayable, memorable, distinct, ownable, on-positioning, context-proof), clear trademark and domain as a gate before anyone falls in love, test finalists in context, and decide with a written reason.

Art-direct a commissioned AI photoshoot end to end, classify the use case (profile picture, print magazine, video ad, commercial, UGC, e-commerce, hero, billboard, thumbnail, food, interior, event), run a full seven-round creative intake, lock a signed-off Shoot Spec Sheet, then write model-ready prompts with camera, lens, aperture, lighting pattern, ratio, palette, and grade specified.

Data-driven bulk image rendering: feed a CMS table or CSV, emit hundreds of on-brand OG cards, ad variants, and listing images from parameterized templates, token-driven and QA'd at scale. Single bespoke assets go to as-design-system / as-ads-meta-creative; this is the at-scale line.

Write, optimize, and test a prompt that works on any model, the general text-prompt craft, distinct from authoring a skill or an image prompt.

Generate branded, scannable QR codes that people actually want to scan. Not boring black-and-white squares. Produces an on-brand QR (brand colors with safe contrast, a center logo, custom eyes, rounded modules) at high error-correction so it stays reliable, with a tracked/short destination URL, exported print-ready (SVG + high-res PNG) at the right quiet-zone and minimum size.

Add real-time graphics to a web page on purpose (ambient grounds, pointer-response fields, 3D objects and materials, globes, environmental scenes, shader effects and physics) each admitted through a four-part test, priced for GPU and weight, shipped with a static fallback and a reduced-motion path.

Make a UI codebase visually consistent and KEEP it that way, audit the component tree for drift (distinct CTA treatments, hardcoded hex, arbitrary radii/type/z-index, nav sprawl), consolidate one primitive at a time into token-backed components, and install a drift ratchet so the numbers only go down.

Turn a reference (a screenshot, a Pinterest board, a live site, a video, or a full-page capture) into a measured design brief carrying an extracted palette with a real contrast matrix, density and rhythm read, corner and depth language, type register, transcribed motion, and a direction contract. Every token anchored to something visible in the reference.

Run the surface-reduction pass on a real app, inventory every screen, nav item, action, field and setting; name one job per screen; measure decisions/steps/fields to value on the top tasks; collapse the information architecture; default instead of asking; defer power to advanced surfaces with a findable path; then publish before/after counts.

Crawl a whole website or web app page by page in a real Chromium browser, capture full-page desktop/mobile screenshots plus DevTools/DOM evidence, force an individual visual review of every captured route and safe UI state, and synthesize one prioritized design-system, UI/UX consistency, responsive, accessibility, and AI-slop audit.

Audit a site, app, screenshot, mockup, or UI codebase for the tells that it was generated rather than made, placeholder copy and names, round fake numbers, filler superlatives, default framework palettes, three-column feature-card grids, flat single-layer shadows, gradient-blob heroes, and reference material carried across without originality. Returns a scored, located, fixable list.

Audit, document, and extend the app's design system (tokens, component variants/states, naming consistency, and hardcoded-value drift) grounded in brand-config/Tailwind/shadcn. Owns design-system documentation + token hygiene; for building premium UI use as-ops-premium-ui-builder, for the WCAG gate use as-core-app-responsive-a11y.

From a product type + audience + stack, select and FREEZE a coherent, industry-grounded design system (style, palette, typography, landing pattern) to a MASTER spec (with per-page overrides) that every downstream builder binds to. Upstream of as-ops-premium-ui-builder.

Plan, run, and synthesize user research for the product, research plans, interview guides, surveys, usability tests, and synthesis into themes/insights/segments tied to a design decision. Owns product/UX research; for sales account research use as-sales-account-research, for market analysis use as-ops-competitor-intel.

Write and review in-product UX microcopy for the Next.js app (buttons, labels, error messages, empty states, CTAs, onboarding, toasts, tooltips) in the Agency Script voice. For marketing/landing copy use as-funnel-copy; for whole-app voice rewrites use as-core-brand-voice-guard; this owns interface strings.

Compile visual shot-language codes into original prompts and produce the picture, exploded views, cyanoprint engineering drawings, turnaround sheets, cutaways, macros, sketchnotes, knolling, schematics, storyboards, and the rest of the mode catalog.

Generate an original, on-brand, reusable VISUAL PROMPT PACK / library for AI image + video generation, master prompts across the styles a brand needs (photography, illustration, 3D/render, avatars, logos/UI graphics, Sora/Veo motion) plus a shot-language mode deck (exploded, cyanoprint, turnaround, knolling, …), each with the brand palette pinned, swappable subject slots.