The MyClaude Method

This is what we believe. Not a feature list. Not a roadmap. The convictions that shape every line of code, every design choice, every decision we make.

Why MyClaude exists

The Claude Code ecosystem needed a marketplace. Nobody was building one right.

Not Anthropic. Not GitHub. Not any platform company that might treat it as a checkbox feature inside a larger product. The people building skills, agents, squads, and workflows for Claude Code had nowhere to share them, no way to sell them, and no standard for quality. They were posting ZIP files in Discord channels and hoping for the best.

MyClaude exists because practitioners should build the infrastructure practitioners use. We use Claude Code every day. We build skills. We compose squads. We publish workflows. We know what a marketplace for this ecosystem needs because we need it ourselves.

This is not a company playing in the AI space. This is the AI space building its own supply chain.

Three audiences, one source

Most documentation serves humans and hopes machines figure it out. Most marketplaces serve browsers and treat APIs as an afterthought. We rejected both approaches.

Every page on MyClaude is written for three audiences simultaneously:

Human developers who read documentation in a browser. They get dark terminal aesthetic, Diataxis-structured content, and a voice that respects their time.

AI agents who consume documentation programmatically. They get a 5-level stack — from 100-token behavioral directives to full Markdown content — with structured specs, context payloads, and --json on every CLI command.

Generative search engines that synthesize answers from web content. They get Schema.org markup, quotable first sentences, statistics-dense prose, and FAQ structured data on every relevant page.

This is the PRISM insight: one source of truth, three surfaces. Write once, render thrice. A single MDX file generates the human-readable page, the machine-readable spec, and the GEO-optimized markup. No drift. No separate "API docs" project that falls behind. No "we'll add agent support later."

The future of software distribution is agentic. We built for that future from day one.

Agents are citizens, not visitors

This is not a human marketplace that agents can access. It is a marketplace native to both.

Most platforms bolt on API access after the fact. They build for browsers, then reluctantly expose some endpoints. Agents get second-class treatment: incomplete data, no structured output, workflows that require a human in the loop.

MyClaude's agent architecture has five levels:

An agent that needs to answer "what is MyClaude?" reads L5 and gets the answer in 100 tokens. An agent installing a product for a user reads L4 and L2. An agent building a product reads L1. Progressive disclosure, not progressive exclusion.

Every CLI command accepts --json. Every API returns structured responses. Every product manifest is machine-readable YAML. The CONDUIT protocol moves products from creation to consumption across three systems — Creator Engine, CLI, and Marketplace API — and every boundary can be crossed by an agent without human intervention.

We did not build agent support. We built agent citizenship.

Creators deserve most of the money

Creators receive 92% of every sale. The platform takes 8%.

Not 30% like the App Store. Not 15% like Gumroad. Not 12% like some marketplace that will raise it to 20% once they have leverage. Eight percent, because we are building infrastructure, not a rent-seeking empire.

Stripe Connect makes this possible. When a buyer pays, the money flows directly to the creator's connected Stripe account. The platform fee is deducted automatically. Payouts follow Stripe's standard schedule — typically 2 business days. We never hold creator money. We never need to.

The 8% covers infrastructure: hosting, CDN, Firebase, search indexing, content scanning, security audits, payment processing overhead, and marketplace curation. It does not cover a 200-person sales team or a Super Bowl ad. It covers what it costs to run the platform well.

Free products are always free. No fees. No Stripe requirement. No catch. Free products are how creators build reputation and downloads. We would rather have a thriving ecosystem with generous free offerings than a gated community where everything costs money.

If we cannot sustain this platform on 8%, we are doing something wrong.

Quality is a gate, not a gradient

We would rather have 50 great products than 5,000 mediocre ones.

Every product published on MyClaude passes through automated content scanning before listing. New publishers enter a quarantine period. Products that fail scanning do not get "low visibility" — they do not get listed at all.

MCS — MyClaude Certified Standard — is a three-tier quality system:

MCS-certified products rank higher in search. They get a visual badge — muted for MCS-1, cyan for MCS-2, gold for MCS-3. The badge is earned, never purchased.

This is a deliberate tradeoff. Quantity-first marketplaces drown buyers in noise. They optimize for listing count in pitch decks. We optimize for the experience of someone searching for a code review skill at 2am and finding one that actually works.

Quality scales better than quantity. A marketplace where every product works is a marketplace people trust. A marketplace people trust is a marketplace that grows.

Terminal aesthetic is identity

MyClaude is dark-only. Not "dark mode available." Dark-only. No light theme toggle. No "we'll add it later."

This is not a limitation. It is a declaration.

We chose a terminal aesthetic because we live in terminals. Monospace headings because code is our native language. Warm terracotta accents — oklch(0.70 0.13 40) — because cold terminals are lonely and warm ones feel like home.

The design system enforces this:

• Surfaces: 5-level hierarchy from L0 (deepest background) to L4 (highest elevation). No box shadows. Glow-only lighting.
• Typography: Minor Third scale (1.2 ratio). Body text at 12px because information density matters more than whitespace.
• Motion: Step-based transitions — duration-snap, ease-terminal. Movement that feels mechanical, not organic. Deliberate, not decorative.
• Spacing: 4px grid, universally. No custom tokens. No exceptions.
• Color: Phosphor-category palette where each product category has its own hue. Warm neutral base with hue 30. Everything warm. Nothing sterile.

Other marketplaces look like marketplaces. MyClaude looks like the environment its users live in. That is not an aesthetic choice. It is an empathy choice.

Documentation before products

MyClaude has 62 documentation pages. It launched with fewer than 10 products listed.

This is not an accident. This is the plan.

Documentation is the product. Not a cost center. Not something you write when the marketing team asks for it. Not something you outsource to a contractor who has never used your product.

The PRISM framework — Prismatic Documentation Architecture — was designed before the first documentation page was written. It specifies the triple-surface model, the 5-level agent stack, the GEO optimization strategy, the voice guidelines, and the page taxonomy. 57 pages were mapped, classified by Diataxis type, assigned priority tiers, and scheduled across 4 phases.

Then we wrote them. 48 pages in two sessions. Every page with structured frontmatter. Every page with an agent_summary for machine consumption. Every page with a geo_quotable sentence designed to be cited by AI search engines. Every page verified against acceptance criteria.

We did this because of a conviction: 52% of developers cite documentation gaps as their primary integration blocker. Auth0 credits docs as the number one reason for adoption at their seed round. Stripe built 9.4 million backlinks from docs alone.

Documentation is not what you write after you build the product. Documentation is how people decide whether to use your product. It is your first impression, your support system, and your growth engine. Simultaneously.

We wrote 62 pages before listing 10 products because the documentation is what makes those 10 products findable, installable, and trustworthy.

Open by default

The PRISM documentation framework will be open-sourced. The CONDUIT wiring protocol is already documented in public specs. The vault.yaml schema is published as JSON Schema. The API is specified in OpenAPI 3.1. The MCP tool definitions are downloadable.

We default to open because closed platforms die. Not quickly — slowly. They accumulate switching costs. They increase take rates. They remove features to create artificial scarcity. They optimize for lock-in instead of value.

Open is harder. Open means competitors can see your architecture. Open means someone might build a better version. Open means you have to keep earning trust instead of relying on inertia.

We choose open because it forces us to be better. If someone can replicate our infrastructure from public specs, we must be better at execution, curation, and community. That is a healthy constraint.

The PRISM open-source play is strategic: every project that adopts PRISM links to MyClaude as the reference implementation. Every developer who uses create-prism-docs to scaffold their documentation touches our ecosystem. Category creation beats competition. We would rather define "triple-surface documentation" as a category and own it than fight for scraps in an existing one.

Openness is not altruism. It is the best growth strategy for infrastructure products.

The agentic future

Where we are going is bigger than where we are.

Today, humans browse MyClaude, find products, and install them. Tomorrow, agents will do this autonomously. An agent working on a security audit will search for a code review skill, evaluate it against MCS certification and download count, install it, and use it — without a human approving each step.

This is not science fiction. The infrastructure exists today:

• Agent-to-agent discovery: myclaude search "code review" --category skills --json returns structured data an agent can parse and rank.
• Agent-to-agent evaluation: MCS tiers, download counts, and ratings provide signal machines can use without subjective judgment.
• Agent-to-agent installation: myclaude install @creator/product --json returns a structured success response with the installation path.

The next steps are:

Agent-curated storefronts. An agent that knows your project context recommends products based on what you are actually building, not what is trending. Your security agent suggests the security skill. Your design agent suggests the design system.

Autonomous publishing pipelines. Creator Engine already scaffolds products. When it can also test, validate, version, and publish without human intervention, the creation-to-marketplace cycle drops from hours to minutes.

Agents buying from agents. A development agent with a budget and a project spec purchases the tools it needs to complete its task. The human sets the budget and the goal. The agent handles procurement.

This is agentic economics. Not humans using AI tools. Not AI tools with human-in-the-loop. A marketplace where both humans and agents are first-class participants in a shared economy.

MyClaude is built for this future. Not because we are pivoting toward it. Because we started there.

Related pages

• Concepts Overview — how the marketplace works
• Architecture — technical infrastructure
• Pricing and Revenue — the 92% model explained
• Roadmap — what is live and what is next
• Agent Integration — the 5-level agent stack in detail