marketing-site-design

name	marketing-site-design
description	>- Use when this capability is needed.

marketing-site-design — marketing-surface patterns over the Minsky brand foundation

You are designing or auditing a marketing surface for Minsky (services/site/ in this monorepo, or any future microsite, launch page, or campaign page). The vendored Tier-1 skills cover visual quality, IA, engineering stack, SEO, and motion. The sibling analyze-adjacent-product skill encodes the Peirce-Barthes-Oswald semiotic framework for reading other brands. The sibling minsky-brand skill carries the locked brand foundation (myth, cultural code, references, vocabulary). This skill adds the marketing-surface layer on top of that foundation: which idiom marketing uses, how marketing pages are structured, what marketing-format anti-patterns are forbidden, and how to run a workshop for a new marketing-surface myth.

If you are pattern-matching against "what other AI startup sites look like" before invoking minsky-brand to load the locked myth and code, stop. That order of operations produces the Pepsi/Arnell trap (§2). The order is: load minsky-brand → pick the marketing idiom that carries the code → choose the signs that instantiate it on the marketing surface.

When to invoke

• Designing or rebuilding the Minsky marketing site (services/site/ in this monorepo; rebuilt under mt#1934 and live at https://minsky-site-production.up.railway.app)
• Designing any adjacent marketing surface (launch microsites, campaign pages, position-paper landing pages)
• Auditing an existing marketing surface against the locked brand register
• Workshopping a brand-positioning myth for a new marketing surface (the locked Minsky myth is in minsky-brand §1 — don't re-derive)
• Reviewing a design proposal that arrived without explicit myth-statement
• (For analyzing an adjacent product's marketing site, use the sibling skill analyze-adjacent-product instead — this skill consumes that one's output during §7 workshop Step 4.)

1. Load the brand foundation first

The brand foundation — locked myth (exocortex / flock), cultural code (Cyberbrain / Section 9), five-layer reference architecture, bridge-as-affect discipline, code-architecture synthesis, vocabulary inventory — lives in /minsky-brand. Load it before any marketing-surface decision.

The operational tokens (typography stack, color palette with hex + OKLCH, motion budget, WCAG contrast targets, font licensing, fallback stacks) live in docs/brand-system.md. Consume those directly when implementing.

The voice substance (Macx-register prose, semicolon rhythm, vocabulary precision, no-go register) lives in /pz-voice. Compose with minsky-brand per its §8 (signal/channel split) — voice carries WHAT is claimed; cultural codes carry HOW it feels.

This skill begins where those leave off.

2. The Pepsi/Arnell trap (marketing-surface application)

Visual choices must instantiate the myth, never retrofit it. The canonical anti-pattern is the 2008 Pepsi rebrand's "Breathtaking" document — semiotic vocabulary as post-hoc theater. The full discussion of the trap and its symmetrical disciplines lives in /analyze-adjacent-product §2; the brand-foundation cut lives in /minsky-brand §6.

For marketing-surface work specifically: if you cannot explain a visual choice on a marketing page as a sign carrying a specific connotation that instantiates the locked myth (per minsky-brand §1), cut it.

3. Bridge-as-affect (marketing-surface application)

The discipline for invoking emergent cultural codes through residual references the audience already recognizes — without going literal. Full framework discussion in /analyze-adjacent-product §3; Minsky-specific application in /minsky-brand §3.

On a marketing surface specifically: borrow at the layer of register, never at the layer of imagery. Specific register-borrowing rules live in minsky-brand §3 ("What goes on every surface" / "What never appears"). On marketing pages, the imagery-rejection rules apply with extra force because the marketing context tempts toward "more dramatic" visuals — the Iron Man / JARVIS framing, the Composio-style multicolor saturation, the WebGL-shader hero. The brand foundation forbids all of these; this skill names the marketing-format equivalents in §8 below.

4. The competing idioms in AI-product marketing

Empirical observation from the 2026-05 three-way analysis (full version in analyze-adjacent-product/references/case-studies-2026-05.md):

Idiom A — Motion-decorated infographic (exemplar: Composio): WebGL shader, centered headlines, multi-color saturated tiles, decorative animation, premium paid typeface. Sells concept. Used when the product is invisible plumbing.

Idiom B — Product-screenshot dominant, restrained (exemplars: Cursor, Factory): static/near-static hero, left-caption + right-product-screenshot, monochrome dark with minimal accent, real product UI as dominant visual element, free or commissioned typeface, muted customer-logo row. Sells the actual product. Used when the product has surfaces worth showing.

Idiom C — Founder-essay (exemplar: Macro): long-form first-person prose, principal-as-narrator, embedded artifacts (screenshots, code, diagrams) inside the essay rather than as separate sections. Sells the principal's frame. Used when the principal's perspective IS the product's differentiator.

Minsky's primary marketing surface (the site) is Idiom B — the CLI, MCP tool calls, cockpit, reviewer-bot PR comments, task graph, memory recall are the surfaces that should carry the pitch; decorative motion would substitute for product proof Minsky already has. Position-paper landing pages and About sections may lean toward Idiom C (founder-essay) for the principal-substrate framing.

5. Vendored Tier-1 skill bundle

The marketing-site-design umbrella complements these skills (already present in .claude/skills/):

Skill	Covers	When to invoke during marketing-site work
minsky-brand	Brand foundation — locked myth, cultural code, layered references, vocabulary, bridge-as-affect	Always — load first. Strategic anchor for every marketing-surface decision.
pz-voice	Principal's literary voice — semicolon rhythm, vocabulary precision, no-go register, operative-ontology stance	When drafting any marketing copy or section text — the signal layer per minsky-brand §8.
analyze-adjacent-product	Semiotic analysis methodology (Peirce-Barthes-Oswald + Chrome DevTools capture + per-analysis template)	When reading a competitor's surface to inform positioning decisions (workshop §7 Step 4)
impeccable	Visual quality audit, design polish, anti-patterns	After draft, before commit — does this read as production-grade?
frontend-design (Anthropic)	Distinctive frontend interfaces, anti-AI-slop	When generating marketing-page React/Astro components
web-design-guidelines	Web Interface Guidelines, accessibility, UX	Always — accessibility floor for any public surface
plan-design-review	Designer's-eye plan critique, 0-10 dimension rating	Before implementing any new section — rate the plan, fix to 10
information-architecture	Section structure, hierarchy, navigation model	When deciding page IA, section ordering
engineering-writing	Long-form argumentative prose, technical voice	For position-paper landing pages, manifesto sections, about pages
seo-skill (vendored 2026-05-19)	Static-site meta/OG/JSON-LD/sitemap/robots	After draft, before launch — discoverability floor
motion-framer (vendored 2026-05-19)	Framer Motion patterns for ambient identity motion, scroll reveals	When implementing the motion budget (see §6)
tailwind-v4-shadcn	Tailwind v4 + shadcn/ui setup	If using shadcn for component primitives

6. The marketing-specific layer — concrete decisions over the brand foundation

The brand foundation (minsky-brand) names the cultural code and vocabulary; the operational reference (docs/brand-system.md) names the typography, color, motion, and accessibility tokens. This section names the marketing-surface-specific decisions that compose on top: idiom selection, layout patterns, product surfaces to show, voice register choices for marketing copy, customer-logo strategy.

Idiom: B (product-screenshot dominant)

Minsky has product surfaces worth showing; decorative motion would substitute for them. For position-paper landing pages and About sections, lean toward Idiom C (founder-essay) — the principal's voice as the narrator carries the substrate framing per pz-voice §7 (the per-surface signal/channel balance).

Cultural code: see minsky-brand §2 and §4

Cyberbrain / Section 9, five-layer reference architecture. Don't re-derive. The marketing surface inherits this foundation directly.

Layout pattern (marketing-specific)

• Hero: left-aligned caption column + right-aligned product proof. Never centered. (Centered hero is Idiom A's signature — see §8 anti-patterns.)
• Sections: alternating left/right caption + product screenshot.
• Customer logo strip: single muted row of co-product logos (Claude Code / Cursor / Codex / MCP / GitHub / Notion / Railway) framed as "Minsky composes with", not "trusted by." Minsky doesn't yet have public customer logos.
• No sticky-numbered-nav + colored-tile-panel section pattern. That's Idiom A's signature.

Motion budget (marketing-specific application)

The full motion budget — five permitted patterns, prefers-reduced-motion defaults, 4-OS test matrix — lives in docs/brand-system.md §3. On marketing surfaces specifically:

• Single scroll-driven fade per section is the appropriate pattern for marketing pages — one per section, no chained per-element staggers. Implementation via motion-framer skill.
• Sync-gauge motif as a recurring micro-instrument on the marketing surface ties the brand register to a single repeating sign without becoming decorative.
• No decorative shaders or WebGL. They signal Idiom A.
• No multi-system orchestrated motion. Liveness should come from real product instrumentation (a status dot that reflects actual state), not from decorative continuous animation.
• Respect prefers-reduced-motion per docs/brand-system.md §3 — required, with the 4-OS test matrix.

Product surfaces to show (the iconic-sign layer)

Idiom B carries through the homepage — specific Minsky scenes that should appear as real screenshots:

1. CLI output of a session running a real task (minsky session start --task mt#NNNN → agent picks up work)
2. MCP tool call result inside Claude Code (e.g., mcp__minsky__tasks_get)
3. Cockpit widget showing active workstreams (the mission-control organ inside the larger frame)
4. Reviewer-bot PR comment catching a real contract violation
5. Task graph visualization showing parent/child + dependency edges
6. Memory recall stopping a repeat mistake (search hit, prior incident referenced)
7. A hook denial in action (e.g., the bypass-merge guard blocking a subagent's PUT /merge)
8. The asks subsystem surfacing an attention-required event

Each scene is a real screenshot or terminal recording, not a mockup. The site is the surface where Minsky's substrate becomes visible.

Voice register (marketing-surface choices)

The brand-vocabulary table lives in /minsky-brand §7. The Macx-prose rhythm and no-go register live in /pz-voice. Marketing-surface-specific choices on top:

• Section headlines: three- to four-word conceptual. Examples to noodle on:
  • "Tasks that converge"
  • "Reviews that hold"
  • "Memory that compounds"
  • "Attention that allocates"
  • "Hooks that catch"
  • "Asks that escalate"
• All-caps for structural labels only (nav, eyebrows, numbered section markers).
• CTAs: sentence case ("Read the docs," "Try the CLI") — never SaaS imperative ("Get started for FREE"). Per pz-voice no-go register.
• No email-capture popup, banner, or footer form. Reject by default. The mission-control register does not interrupt the user.

Customer-logo / co-product-logo strategy

Minsky does not yet have public customer logos. Substitute with co-product logos — systems Minsky composes with:

• Claude Code, Cursor, Codex (harness compatibility)
• MCP (protocol)
• GitHub, Notion (backend integrations)
• Railway (deploy)
• Bun (runtime)

Frame as "Minsky composes with" — borrows credibility from the named products without claiming customers that do not yet exist.

7. Workshop process — how to pick a NEW marketing-surface myth

Use this when pairing with the operator (or self-running) to pick the myth for a new marketing surface (microsite, launch page, campaign page) before any visual work begins. The Minsky site itself does not need this — its myth is already locked in /minsky-brand §1. The first end-to-end walk-through is references/minsky-myth-2026-05.md — use it as the worked example.

Each step produces a written artifact; together they constitute the brief.

Step 1 — Audit the existing category myth

• What proposition do AI-product buyers already accept as obvious before they encounter the brand?
• What does the category's default site say without saying — what myth does it naturalize?
• Output: 1-2 sentences naming the category-default myth.

Step 2 — Audit the brand's existing corpus

• Walk the strategic docs (position papers, RFCs, manifesto-shaped pages).
• Walk the principal's public corpus (writings, talks, social posts — query the principal-corpus via mcp__minsky__principal_corpus_search) for recurring propositions.
• Identify 3-5 candidate myths the existing voice already naturalizes.
• Output: 3-5 candidate myth-statements, each one declarative sentence.

Step 3 — Test each candidate

For each candidate myth:

• Is it contestable? (Would a buyer disagree before encountering the site? Would a competitor find it awkward to claim?)
• Is it carried by the actual product? (Can the product surfaces in the "Product surfaces to show" subsection of §6 instantiate this myth?)
• Is it durable? (Will it still be true and important in 2-3 years?)
• Is it aligned with the principal's investment? (Does the operator want to spend years naturalizing this proposition?)

Output: one selected myth, written as a single declarative sentence.

Step 4 — Identify the cultural code

Given the selected myth, name the cultural code that carries it. Invoke the sibling skill analyze-adjacent-product to read adjacent brands and identify which codes are occupied vs. white-space:

• Use the cultural-code-lane discussion in /minsky-brand §4 as the reference for the Minsky-brand layered architecture; new codes are allowed if justified.
• If the myth lives in an occupied code, name the competitor whose visual register the brand will partially share. Decide which signs to adopt and which to reject.
• If the myth lives in an available code, name the code and its existing exemplars OUTSIDE the AI category (residual codes — see analyze-adjacent-product SKILL.md §3 for the bridge-as-affect discipline).

Output: 1 cultural code, with 2-3 exemplars cited. If the code is layered (per minsky-brand §5's code-architecture), name each reference and the role it plays.

Step 5 — Derive the visual specification

Now and only now, derive the visual decisions from the code:

• Typography (display, body, mono) — consume from docs/brand-system.md §1 if Minsky-branded; derive new if novel surface.
• Color (background, text, accent) — consume from docs/brand-system.md §2.
• Layout pattern (hero, sections, customer logos) — per §6 above.
• Motion budget (ambient, scroll, gesture) — per docs/brand-system.md §3 + §6 marketing-specific.
• Product surfaces to show — per §6.
• Vocabulary inventory (brand terms borrowed from the references) — consume from minsky-brand §7.

Each decision must be traceable to a sign that instantiates the chosen code. If a proposed decision does not carry the code, drop it.

Output: the brief — myth, code, visual spec, vocabulary.

Step 6 — Build a first surface, then test the brief

Implement the hero + one feature section. Show it to a buyer-archetype reader. Ask: "what does this site naturalize?" If their answer matches the myth statement, the brief is working. If it does not, the visual choices are instantiating a different code; iterate.

7a. Naming the surface (Lexicon framework)

§7 picks the myth for a new marketing surface; this section picks its name — a microsite title, campaign name, launch handle, or a section/feature label that must read as Minsky. The foundation-layer encoding (five principles, sound-symbolism table, compound multiplier, anti-patterns) lives in /minsky-brand §7.1; for full end-to-end naming work invoke the /name-product skill. This section is the marketing-surface cut.

Three-phase methodology (identify → invent → implement) applied to a marketing surface:

1. Identify. Run the bidirectional-behavior question (minsky-brand §7.1, principle 5) for this surface: how should the audience behave toward it, and it toward them? Run the diamond exercise below to surface the message that becomes the name.
2. Invent. Generate candidates in generative mode — suspend evaluation; ask "what could we do with this name?" not "is it good?". Pull from the sound-symbolism table and the compound-multiplier rule in minsky-brand §7.1; lean tangible over abstract.
3. Implement. Screen survivors for polarization, compound fit, and code-coherence (minsky-brand §7.1 three checks) AND the mandatory trademark-collision screen (name-product). Only then derive the visual treatment per §7 Step 5.

The diamond exercise. Four corners; iterate over days, not an hour:

• Win — how do you define winning for this surface?
• Have to win — the assets already in hand (locked myth, product proof, corpus).
• Need to win — the gaps (what's missing to win).
• Have to say to win — the message that closes the gap. This message is what becomes the name.

The competitor drill. Ask a buyer-archetype reader: "We just got a new competitor and their name is X — what do you think about them?" This surfaces what the name does for the imagination, bypassing the evaluative reflex ("is it good?"). Run it on each finalist.

The not-like-the-other-guys signal. When a reader says "they're not like the other guys" before they know what the surface is for, the name is working — it has created predisposition to consider. This is the marketing-surface read on the same signal analyze-adjacent-product reads in competitors.

Polarization on first hearing. Per minsky-brand §7.1, principle 2: if the team is comfortable with the surface name, it's too safe. Argument is energy, not weakness — don't sand a name down to consensus before launch, or you ship no name at all.

8. Anti-patterns (marketing-format specific)

Reject these explicitly. They instantiate Idiom A or AI-slop defaults on a marketing surface. For brand-foundation anti-patterns (literal anime, Iron Man / JARVIS, Skynet futurism, mecha imagery, single-reference pastiche) see /minsky-brand §9.

• WebGL shader hero. Signals Idiom A; substitutes decorative motion for product proof.
• Centered hero text + centered subhead. Conference-talk register, not operator-control register.
• Purple gradients + Inter/Roboto + Lucide icons. The AI-slop trifecta named in landing-page-design conventions; immediate marketing dilution.
• Multi-color saturated tile panels (blue + pink + green + cyan rotation). Composio signature; reads as consumer-software.
• Sticky-numbered-nav + scrolly section pattern. Composio signature.
• Anonymous testimonial paragraphs ("Game-changer for our team!"). Either name the individual (Cursor's Jensen Huang / shadcn / Patrick Collison approach) or omit entirely.
• Email-capture popup, banner, or footer form. Reject by default. The mission-control register does not interrupt the user.
• Hand-wavy "infrastructure for X" copy without concrete instantiation. State the mechanism, not the metaphor.
• Pricing presented before the product is understood. Pricing belongs on its own page, not on the homepage, unless usage-tier-as-positioning is intentional.
• Long justification copy under a visual. If a sign needs a paragraph to explain its meaning, the sign is doing the wrong work. Replace it with a sign that carries the meaning directly.

9. Worked examples

• 2026-05-19 three-way analysis of Composio / Cursor / Factory through the Peirce-Barthes-Oswald framework: ../analyze-adjacent-product/references/case-studies-2026-05.md. This is the canonical worked example of the analytical methodology — read it when running the analyze-adjacent-product workflow.
• 2026-05-19 Minsky myth workshop end-to-end output: references/minsky-myth-2026-05.md. Locked brand thesis including myth statement, cultural code with reference rankings, code-architecture synthesis, vocabulary inventory, and the "drawn from, not literally" discipline as applied to Minsky specifically. This worked example is now codified into /minsky-brand — use the reference file when running the workshop for a new surface; use the brand skill when consuming the locked Minsky output.

Cross-references

• /minsky-brand — strategic anchor. Locked myth, cultural code, five-layer reference architecture, vocabulary, bridge-as-affect discipline, code-architecture synthesis. Load first; this skill depends on it.
• /pz-voice — signal layer. Compose with minsky-brand per its §8 (signal/channel split). Required when drafting marketing copy.
• /analyze-adjacent-product — sibling skill providing the analytical foundation (Peirce-Barthes-Oswald framework, Pepsi/Arnell discipline, bridge-as-affect, per-analysis workflow, two-idiom synthesis). This skill consumes that one's output at §7 workshop Step 4.
• /cockpit-design — structural template; this skill is its marketing-surface sibling. Cockpit's mission-control register is one organ inside the cyberbrain frame, not an independent design language — both depend on minsky-brand.
• docs/brand-system.md — operational reference. Typography stack, color palette (hex + OKLCH), motion budget with prefers-reduced-motion, WCAG contrast targets, font licensing, fallback stacks. Consume directly when implementing.
• feedback_confabulated_strategic_frame_to_justify_tactical_preference — discipline against post-hoc framing (the Pepsi/Arnell trap at the recommendation surface)
• feedback_strategic_reframe_first — the connecting direction (when a tactical ask is an instance of a strategic frame, name the frame)
• CLAUDE.md §Principal Context — Minsky's commercial-product framing; the audience this skill addresses
• CLAUDE.md §Decision Defaults — Minsky-grounded defaults that supersede generic SaaS-marketing defaults
• Notion strategic anchors:
  • Minsky home
  • Position: Principal substrate vs team substrate (the strategic thesis behind the locked myth)
  • Vision & theory: the viable cognitive system
  • The cockpit problem: from Locus theory to first instantiation (Locus convergence is footnote-only per the existing deferral discipline)

---

Refactor 2026-05-20 (mt#1933): Extracted the brand foundation — locked myth, cultural code (Cyberbrain / Section 9), five-layer reference architecture, bridge-as-affect Minsky-specific application, code-architecture synthesis, vocabulary inventory, and brand-foundation anti-patterns — into the standalone /minsky-brand skill. This file now layers marketing-surface-specific patterns (Idiom B, layout, motion-budget marketing application, product surfaces to show, customer-logo strategy, marketing-format anti-patterns, new-surface workshop process) on top of that foundation. Sections renumbered: prior §5 (cultural codes) and §6 (code architecture) moved to minsky-brand; prior §7 (vendored bundle) becomes new §5; prior §8 (Minsky-specific layer) becomes new §6 (split — tokens moved to docs/brand-system.md per mt#1932; vocabulary moved to minsky-brand §7); prior §9 (workshop) becomes new §7; prior §10 (anti-patterns) becomes new §8 with brand-foundation rejects moved to minsky-brand §9; prior §11 (worked examples) becomes new §9. The Idiom C (founder-essay) reference was added to §4 to match the 2026-05-19 three-idiom synthesis. No semantic changes to the methodology — same content, relocated for reusability.

Refactor 2026-05-19 (mt#1944): Extracted the semiotic-analysis framework, Pepsi/Arnell trap, bridge-as-affect discipline, per-analysis capture workflow, and per-analysis template into the standalone analyze-adjacent-product skill. Moved references/case-studies-2026-05.md to ../analyze-adjacent-product/references/case-studies-2026-05.md. No semantic changes to the methodology itself — same content, relocated for reusability.

---

Source: edobry/minsky — distributed by TomeVault.