mbb-page-maker

name: mbb-page-maker description: Create, edit, package, or export consulting-style HTML slide decks for strategy, board, investment, pitch, and executive presentations. Use when the user asks for an HTML PPT, slide deck, pitch deck, investment memo deck, board report, static browser-openable presentation, WeChat-shareable PDF/PNG previews, or updates to an existing MBB Page Maker deck. license: MIT compatibility: Agent Skills-compatible coding agents. Claude Code requires the skill to be visible under .claude/skills or ~/.claude/skills; use references/agent-compatibility.md if an agent reports Unknown skill. metadata: author: "Saitop at NomiciAI saitop@nomici.ai"

MBB Page Maker

This AgentSkill produces static HTML presentation decks with consulting-grade page logic: clear titles, tight storylines, exhibit-first layouts, and executive-ready visual hierarchy.

The source deck must run without a build step: plain HTML, CSS, and JavaScript. Source decks use predownloaded local fonts through assets/css/fonts.css. The delivered package must be self-contained: no remote runtime scripts, remote images, remote font URLs, stylesheet imports, or non-embedded media.

Authoring Philosophy

Treat every deck as a decision tool, not a styled page collection. The agent owns the authoring judgment: infer the audience, decision, storyline, layout, component, and theme from the user's material instead of asking the user to design the deck.

Hard rules:

Start decision-backwards: audience, decision, answer, evidence, implication, action.
Keep source fidelity: use user material, marked assumptions, or approved external data; never invent evidence.
Let evidence shape drive components; if evidence is sparse, use qualitative pages and visible assumptions.
Use answer-first titles and visuals only when they support the argument.
Treat archetypes and showcases as references, not templates; complete the deck and exports when material is sufficient.

Install

Use the copyable terminal command:

npx skills add https://github.com/NomiciAI/mbb-page-maker

Repository: NomiciAI/mbb-page-maker

If an agent reports "Unknown skill" after installation, read references/agent-compatibility.md. Some clients install the canonical package under .agents/skills/ but require a symlink or copy under their own skill directory.

Golden Path

For real deck work, follow this path in order:

Read references/authoring-guide.md before creating or materially rewriting a deck.
For strategy, board, investment, pitch, transformation, or full-deck work, read references/consulting-thinking.md.
Infer audience, decision, likely answer, constraints, and external-data permission.
Extract evidence: claims, numbers, comparisons, timing, risks, decisions, assumptions, and gaps.
Choose references only as needed: references/full-decks.md for full storylines, references/content-to-exhibit-router.md for evidence-to-component routing, templates/showcase/README.md for page patterns, references/visual-assets.md for visual support.
Start from templates/starter-deck.html or the user's existing deck. Do not start from a blank file.
Create a compact internal slide plan: message, evidence source, evidence shape, selected component, selected showcase pattern when useful, fallback, and output role.
Assemble slides using existing CSS layers, layout shells, components, and theme tokens.
Run scripts/check-deck-quality.sh path/to/index.html and scripts/check-deck-contrast.sh path/to/index.html; fix failures.
Run scripts/render.sh path/to/index.html unless the user explicitly asks for source HTML only.
Deliver dist/package/index.html, dist/package/index.pdf, dist/index.pdf, and dist/png/ as the default share set.

If any export step fails, fix the source deck or dependency problem and rerun the same command. Do not hand-wave missing package assets, PDF, PNGs, or audit failures.

Current Scope

This is the foundational HTML skeleton phase. Use the starter deck as the base page system, then expand case-by-case references later.

Use templates/starter-deck.html as the minimal HTML PPT starting point.
Use templates/deck.html only as the design-system gallery and review tour.
Use templates/full-decks/*/README.md as the fast indexing layer, then templates/full-decks/*/index.html as first-class deck archetypes for storyline pacing, page roles, density, component coverage, and export packaging. Do not copy their page order or content verbatim into user decks.
Treat templates/full-decks/*/index.html as agent-facing archetype references. examples/* are independent public demo decks; do not use examples as the authoring source unless explicitly recovering newer content.
Use templates/showcase/README.md and templates/showcase/*.html for reusable page-level thinking patterns and theme + layout + component combinations, not full storyline templates.
Use the CSS layers in order: fonts.css, base.css, layouts.css, components.css, illustrations.css, then one file from assets/themes/.
Use assets/js/runtime.js for keyboard navigation and print/export mode.
Read references/authoring-guide.md before creating a real deck.
Read references/agent-compatibility.md, references/consulting-thinking.md, references/pattern-index.md, references/content-to-exhibit-router.md, references/layouts.md, references/components.md, references/themes.md, references/full-decks.md, references/adding-patterns.md, references/visual-assets.md, or references/asset-sourcing.md only when that catalog is needed.
Run scripts/check-deck-quality.sh path/to/deck.html and scripts/check-deck-contrast.sh path/to/deck.html before final delivery.
Unless the user asks for HTML only, run scripts/render.sh path/to/deck.html and deliver the self-contained package index.html, PDF, and PNG slide images.
Use templates/layouts/default-*.html for opening, centered transition, centered message, headline metric, and ending pages.
Use templates/layouts/blank-*.html for ordinary structure-first pages before adding specialized components.
Normalize relative asset paths after copying snippets into a deck.
Do not add CDN scripts, remote images, dynamic module loaders, or runtime dependencies that cannot be embedded into the final package. Use only the predownloaded local font files referenced by assets/css/fonts.css.
Do not bundle third-party original images by default. Network images may be used only as reference, seed, or mood input unless the user approves a reviewed CC0/public-domain asset. Final bundled visual assets must be self-generated or repo-native, strongly transformed, and free of visible logos, people, trademarks, product UI, source marks, or consulting-firm identifiers.
Do not copy proprietary logos, company names, confidentiality statements, or source-identifying marks from reference decks into generated HTML.
Do not add MBB Page Maker, archetype, agency, or placeholder brand marks to user decks. Use no visible logo unless the user supplies one.
Keep footers minimal: page number and, if useful, a neutral deck title. Do not include "source", "prepared by", consulting-firm-style, or archetype footer copy unless the user explicitly supplied that text.

Design System Layers

assets/css/base.css: slide canvas, typography primitives, light/dark mode tokens, runtime controls, print rules.
assets/css/layouts.css: page-level layout shells.
assets/css/components.css: reusable content components inside layouts.
assets/css/illustrations.css: neutral CSS/SVG-friendly illustration primitives and asset slots.
assets/themes/*.css: color tokens only. Keep theme names neutral.

Composition Workflow

Start from templates/starter-deck.html for generated decks; use skeleton files only as role-specific references.
Infer the communication task from the user's prompt and source material.
For strategy, board, investment, pitch, transformation, or full-deck work, apply references/consulting-thinking.md.
For full decks, read references/full-decks.md, the closest full-deck README sidecar, and then inspect that index.html as an archetype reference.
For single-page or partial-deck combinations, read references/pattern-index.md, references/content-to-exhibit-router.md, or templates/showcase/README.md before composing; adapt and recombine patterns based on the user's evidence.
Apply the consulting authoring gate. If the user gives enough material, do not ask them to design pages; infer the audience, decision scenario, output length, default theme, relevant archetype, optional showcase pattern, and evidence routes. Route LOP, pitch, initiative recommendation, market entry, pricing, org design, diligence findings, PMI, turnaround, and portfolio allocation to their dedicated archetypes before falling back to generic strategy or investment sequences. Ask only when the audience, decision, hard constraints, or permission to use external data are genuinely missing.
Read references/content-to-exhibit-router.md and references/components.md, then create a slide plan before writing HTML. Each planned slide needs: message title, evidence source, evidence shape, selected component(s), selected showcase pattern when useful, fallback, and output role.
Write answer-first slide titles and build a coherent storyline before polishing visuals.
Choose one page layout from templates/layouts/ using references/layouts.md.
Choose reusable components from templates/components/ using references/components.md.
Apply one theme file from assets/themes/.
Verify that the assembled slide has one clear message, one dominant communication structure, no visible overflow, and no copied source identifiers.
Run the deck quality and contrast audits; fix empty section pages, missing evidence components, and failed text/background pairs.
Render the default self-contained HTML package, PDF, and PNG slide images unless the user explicitly asked for HTML only.

Prefer editing the deck and rerunning deterministic scripts over explaining how the user could do it manually.

Do not create filler pages. If the user input does not contain enough structured content for a specialized layout, use a simpler layout or omit the page.

Do not bury structured data in prose. If the source material contains usable numbers, comparisons, rankings, time phases, risks, decisions, or categories, route them to an appropriate component. A data-rich section should not become only cards or a section divider.

Do not create fake data exhibits. If the input lacks numeric values, categories, dates, or comparable records, choose a qualitative component such as outcome-support, framework-map, issue-tree, cause-effect, numbered-list-grid, callout, or decision-log. Never invent numbers to justify a chart, table, matrix, or metric page.

Do not automatically browse for external data. Use only the user's material unless the user explicitly asks for current/public/online research or grants permission after you ask. If external data is approved, keep it separate from user-provided material in the slide plan and use it only where it materially improves the argument.

Do not ask users to choose layouts, components, or page sequences. Ask only for audience, decision, hard constraints, permission to use external data, or missing fields required for a promised exhibit. If the user does not answer, continue with a clear default and mark gaps as assumptions or open questions.

Delivery must fail fast instead of shipping partial files. The package exporter verifies that dist/package/index.html has no external stylesheets, external scripts, or non-embedded media resources. If that check fails, replace the dependency with static HTML/CSS/SVG, local media that can be inlined, or a built-in component before delivery.

Avoid pure section-divider pages in short generated decks. When the user asks for only a few sections, either omit dividers or turn them into useful section-intro pages with 2-4 data signals, a catalog, or a summary component. A divider is acceptable only when it creates needed pacing in a longer deck.

If a pure divider is deliberately needed, mark it with data-allow-divider="true" and keep it out of short evidence-driven decks.

Avoid generic slide titles such as "Overview", "Analysis", or "Findings" unless they are section labels. Content slide titles should state the page's message as a complete, meaningful sentence.

If content overflows a region, reduce the number of components, choose a wider layout, or split the content into another slide. Do not solve overflow by making text too small to read.

If a slide has one compact component or compact region layout that only uses roughly the top half of the content area, vertically center it with .content.is-vertically-centered or .content[data-align-y="center"]. This also applies to two-column body slides when both sides are real primary content but both are short: for example a 3-5 row table next to a 3-4 row ranked bar chart, two compact scorecards, a small roadmap next to a short implication stack, or paired callouts. For sparse individual regions, use .region-body.is-centered. If you create a custom split wrapper inside a vertically centered content area, do not leave that wrapper or its direct child stacks at height: 100%; use height: auto, .is-fit-content, or data-fit-height="content" so the outer content area can actually center the exhibit. Keep dense tables, tall timelines, multi-section dashboards, and pages with deliberate top/bottom staging top-aligned. Do not leave a short table, roadmap, timeline, flow, or right-rail layout pinned to the top with a large empty lower half.

Dark or pitch-style pages must use .dark, [data-mode="dark"], [data-tone="dark"], .dark-cover, or [data-variant="dark-cover"] instead of only setting a dark background. The runtime includes a slide-level safety net for dark backgrounds, but generated decks must still pass the contrast audit.

Page choice is compositional, not fixed. Infer the user's task and data shape, then combine the smallest suitable layout with the needed components. Full-deck archetypes and showcase patterns are references, not constraints; diverge from them whenever the user's material calls for a better storyline or exhibit. For input text, PDFs, tables, transcripts, or scraped material, first build an evidence inventory and route it through references/content-to-exhibit-router.md; for flows, funnels, filters, horizons, timelines, journeys, tables, lists, and quotes, inspect the relevant showcase before inventing ad hoc HTML. Use references/layouts.md and references/components.md for detailed selection guidance.

Visual choice is also compositional. Use visuals only when they support the answer-first message, improve visual balance, or create a deliberate cover/section moment. Do not add illustration to data-rich pages by default; charts, tables, scorecards, matrices, metrics, and risk/status components remain primary when evidence exists. Visual primitives and generated images are not evidence and do not count as the page's content component by themselves. Never use abstract visuals to fill empty space on a body slide; if the visual can be removed without weakening the argument, remove it. Before using bitmap media or interior visual primitives, inspect references/visual-assets.md and templates/showcase/visual-patterns.html.

Do not force a specialized page. If the input is sparse, use a simpler page or ask for the missing data.

Content sufficiency examples:

Contact roster: require at least 3 contacts with names plus at least one of role, location, email, or team. Principal/contributor grouping requires an explicit group field or clear wording.
Centered metric: require one verified numeric value and one sentence explaining the business meaning.
Chart: require numeric values with labels. Do not invent data.
Table: require comparable rows and columns. If only prose exists, use cards or a statement page instead.
Timeline/roadmap: require phases, dates, or ordered steps.
Matrix: require two meaningful dimensions.

Default theme selection:

Honor an explicit dark, light, or mixed style request across the whole deck. If style is unspecified, choose one theme and one dominant tone; ordinary analytical body slides should default to light or neutral mode, with dark mode reserved for covers, closings, deliberate section dividers, or a few high-emphasis story moments. Do not alternate dark and light body slides merely to make the deck feel more dynamic.
mono.css: structure-first drafts and neutral skeletons.
blue.css: default executive consulting style.
red.css: urgency, turnaround, risk, commercial action.
green.css: growth, sustainability, operations, transformation.
pitch.css: investor, board, fundraising, and VC-style pitch decks. Treat it as a cohesive pitch token set, not a requirement to mix dark and light slides throughout the deck. For institutional finance, infrastructure, strategy, or board materials without an explicit fundraising or marketing-style ask, prefer blue.css or mono.css with light analytical body pages.

HTML Slide Contract

Decks should be composed from self-contained slide sections:

<main class="deck" data-deck-title="Market entry strategy">
  <section class="slide cover" data-title="Market entry strategy">
    ...
  </section>
  <section class="slide" data-title="The market is attractive but access is concentrated">
    <header>...</header>
    <div class="content">...</div>
    <footer class="footer">...</footer>
  </section>
</main>

Ordinary slides use header + .content + footer. Cover, ending, and full-bleed slides may use deliberate variants. Insert components only into .content, .region-body, .safe-fill, or .safe-stack.

Default export targets: self-contained HTML package, PDF, and PNG slide images. If the user does not specify an output format, create all three by running scripts/render.sh path/to/index.html; when package and PDF export are both enabled, the generated PDF is copied to dist/package/index.pdf for easier handoff. Future export target: PowerPoint after the HTML system is stable.