- name
SKILL.md
技能檔案保留在發布者的儲存庫中。
在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add notque/vexjoy-agent --skill frontend-slides -a cursor -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add notque/vexjoy-agent --skill frontend-slides -a claude-code -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add notque/vexjoy-agent --skill frontend-slides -a codex -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add notque/vexjoy-agent --skill frontend-slides -a grok -y在 Grok Bot 聊天中傳送複製的訊息。Bot 會從 GitHub 取得這個 SKILL.md,並加入你所有 Bot 共用的技能中。
Install this skill: https://github.com/notque/vexjoy-agent/tree/HEAD/skills/frontend/frontend-slides來自發布者
Generate browser-based HTML presentations as a single self-contained .html file. Three entry paths: new build from topic/notes, PPTX-to-HTML conversion, or enhancement of an existing HTML deck.
Routing disambiguation: When the user says only "slides" or "deck" without specifying format, ask exactly one question before proceeding: "Should this be an HTML file (opens in browser) or a PowerPoint file (.pptx)?" This skill only produces HTML — it converts an existing PPTX to HTML but cannot generate a new .pptx file.
Assemble CSS: python3 skills/frontend/frontend-slides/scripts/assemble-styles.py --preset <name>. Use --list for all 12, --mood <word> to match.
Templates: templates/base.css (mandatory base), templates/presets/*.css (preset :root variables).
Assemble layout CSS: python3 skills/frontend/frontend-slides/scripts/assemble-layouts.py --layouts title,content,code. --all for everything, --include-html for HTML templates.
Each layout in templates/layouts/ has paired .html + .css: title, content, grid, code, quote, image, section-break.
Assemble controller: python3 skills/frontend/frontend-slides/scripts/assemble-controllers.py --core. Add --features speaker-notes,countdown-timer.
Identify which of the three paths applies:
GATE 1: Do not proceed without identifying the path. If the input is ambiguous (e.g., "make me a deck" with no file and no topic), ask one question to resolve it.
Ask exactly three questions -- no more:
Collect or generate the content before touching any style decisions.
GATE 2: Content exists (or the user has confirmed topic-only intent with an explicit "I'll provide content as we go") before Phase 3 begins. Do not start style work with no content direction.
Use the Style Presets table (above) to map mood to preset names.
Two sub-paths:
Sub-path A -- User names a preset directly: Skip previews. Confirm the preset name exists in the Style Presets table. Proceed to Phase 4.
Sub-path B -- User does not know the preset: Ask for mood using exactly these four options: impressed / energized / focused / inspired. Map the mood to candidate presets using the Style Presets table -- translate the user's mood description to a preset name rather than asking them to choose from a list. Generate 3 single-slide HTML preview files in .design/slide-previews/ -- one per candidate preset -- using real slide content (not placeholder lorem ipsum). Present the previews and ask the user to pick.
GATE 3: User has either named a preset from the Style Presets table or selected one of the three previews. A vague direction like "make it look professional" is not sufficient -- a named preset must be confirmed before Phase 4. If no selection is made, regenerate previews with different presets. Never fall back to a generic purple gradient; presets exist to avoid exactly that.
Build the presentation as a single .html file with all CSS and JS inline (no external CDN dependencies). Follow these rules:
CSS base block verbatim: Copy the mandatory CSS base block from templates/base.css exactly as written -- do not paraphrase or rewrite it. Apply the chosen preset's theme variables on top. The base block must be present character-for-character because the validation script checks for it.
Viewport fit on every slide: Every .slide element must have height: 100vh; height: 100dvh; overflow: hidden. When content overflows, split the slide into multiple slides -- never shrink text, add scrollbars, or set min-height that could allow growth past 100dvh. A slide with scrollable content is a web page, not a slide.
Density limits: One idea per slide, one headline that wins. Maximum 6 bullets per content slide and about 12 words per bullet. If content needs a 7th bullet, split into two slides -- dense text is unreadable in presentation context.
Responsive sizing: All body text must use clamp() for font sizing. No fixed-height content boxes (height: 300px on inner elements). For images or code blocks that need height constraints, use max-height: min(Xvh, Ypx) with overflow: hidden.
CSS negation rule: Never write -clamp(...) -- browsers silently compute it to 0, causing text to disappear. Always write calc(-1 * clamp(...)) when a negative value is needed.
JS controller class: Implement SlideController with all of the following (not optional):
ArrowRight/ArrowLeft/Space forward; ArrowLeft/Backspace backward; Home/End for first/lasttouchstart/touchend with 50px thresholdwheel event (150ms) -- without debounce, wheel events cause multi-slide jumps. Add a navigating flag that blocks re-entry during transition.currentSlide / totalSlides visible in corner.visible class when slide enters viewport for reveal animations. Use opacity: 0 + transform: translateY(20px) to hide slides before reveal -- never display: none, which prevents Intersection Observer callbacks.Optional features (off by default, add only when user requests):
n key@media print CSS for PDF-via-browser exportGATE 4: The output HTML file exists on disk and contains the verbatim mandatory CSS base block from templates/base.css. Verify with a string search before proceeding -- a file that "looks right" is not sufficient.
Run the deterministic validation script:
python3 skills/frontend/frontend-slides/scripts/validate-slides.py path/to/output.html
Exit codes:
0 -- All slides pass at all 9 breakpoints. Proceed to Phase 6.1 -- Overflow detected. The script prints which slides overflow at which breakpoints. Fix by splitting the overflowing slides. Re-run validation. Do not proceed until exit code is 0. Content that fits at 1920x1080 but overflows at 375x667 still fails -- clamp() sizing solves most cases; if not, split the slide.2 -- Playwright unavailable. Fall back to the manual checklist gate below. Tell the user validation is running in manual mode and is less reliable.Manual checklist gate (fallback, only when exit code is 2):
For every slide, verify all of the following. If any item fails, fix it before proceeding.
height: 100vh and height: 100dvh present on .slideoverflow: hidden present on .slideclamp() for font sizingheight: 300px on inner elements)min-height on .slide that could allow growth past 100dvh-clamp(...) patterns anywhere in CSSLook, then fix (final check): Run the core doc's section 10 loop on the title slide, the densest content slide, and one code or chart slide, at 1280x720 and 375x667. Each slide should pass the squint and grayscale tests with one clear headline. Fix, re-render, and stop after about three rounds.
GATE 5: Exit code 0 from the validation script, or -- only if Playwright is unavailable (exit code 2) -- explicit user confirmation that the manual checklist passed for every slide. User confirmation must enumerate the slide count checked. The look-then-fix loop has run, or the user has the checklist when no renderer exists.
Delete .design/slide-previews/ unless the user explicitly asked to keep them:
rm -rf .design/slide-previews/
Open the file with the OS-appropriate command:
open path/to/output.htmlxdg-open path/to/output.htmlstart path/to/output.htmlPrint a delivery summary:
File: path/to/output.html
Preset: [preset-name]
Slides: [N]
Theme: [3 CSS custom properties easiest to change for re-theming]
GATE 6: Delivery summary printed. File exists at the stated path. Previews deleted (or user confirmed to keep). Task is complete only when all three conditions are met.
Browser-based HTML presentation generation.
---name: frontend-slidesdescription: Browser-based HTML presentation generation.---# skills/frontend/frontend-slides/SKILL.md常見問題
從發布者下載 skills/frontend/frontend-slides/SKILL.md,或複製路徑後開啟儲存庫。PlugMCP 只收錄這項技能,不代管檔案。
發布者將它存放在 https://github.com/notque/vexjoy-agent。
Claude、Cursor、Codex 和 Grok Build 可以讀取你從發布者儲存庫加入的 SKILL.md。在聊天中傳送技能的 GitHub 資料夾連結,Grok Bot 就會安裝它。
Reduced-motion: Include a @media (prefers-reduced-motion: reduce) block that suppresses all transitions and animations.
Font loading: Use @font-face with font-display: swap and a system font stack fallback. Never reference external CDN fonts without a local fallback. Missing font-display: swap causes invisible text during load (FOIT).
PPTX path: If converting from PPTX, use python-pptx to extract text, notes, and asset paths. Preserve slide notes and maintain asset order. If python-pptx is unavailable, print a clear error and ask the user to install it (pip install python-pptx) or provide content manually -- do not silently skip content.