Linkedin Carousel Generator

LinkedIn Carousel Generator

Generates a 13-slide LinkedIn carousel — LinkedIn calls this a "document post" — that visually walks a reader through what makes an intelligent textbook worth exploring. Unlike linkedin-announcement-generator (which writes the post's caption text), this skill produces the artifact itself: a square, mobile-readable slide deck the reader swipes through in their feed, built from the project's real metrics, mascot art, and screenshots.

When to Use This Skill

Use this skill when the user says things like:

• "Make a LinkedIn carousel/slideshow for this textbook"
• "Create a LinkedIn document post"
• "I want a slide deck I can upload to LinkedIn"

Do not use this skill when the user wants only the caption/post text — use linkedin-announcement-generator for that (the two are often used together: carousel as the attached document, announcement text as the post copy, with the carousel referenced instead of a bare link).

Do not use textbook-to-presentation-generator for this — that skill builds a 25-35 slide lecture deck for live classroom delivery with speaker notes. This skill builds a much shorter, image-heavy, no-speaker-notes deck meant to be read silently on a phone in a feed.

Prerequisites

• Node.js installed, with pptxgenjs installed locally in the output directory (npm install pptxgenjs)
• An intelligent textbook project with:
  • mkdocs.yml (site name, description, theme palette)
  • docs/about.md (the "Why This Book" narrative)
  • docs/learning-graph/book-metrics.json (canonical metrics — see book-installer's bk-generate-book-metrics if missing)
  • docs/img/mascot/ with the character sheet and pose PNGs (see book-installer's learning-mascot guide if the project has no mascot yet)
  • docs/license.md and docs/img/license.png
  • At least one MicroSim under docs/sims/, and (ideally) a learning-graph viewer MicroSim (commonly docs/sims/graph-viewer/)
• ~/.local/bin/bk-capture-screenshot available (installed by microsim-utils) for the two screenshots this skill needs
• Python 3 with Pillow (pip install pillow) for cropping the learning-graph screenshot

If any prerequisite is missing, tell the user which one and point at the skill that creates it rather than improvising a substitute (e.g. don't hand-draw a learning graph image if book-metrics.json and a real graph viewer exist).

The 13 Slides

This mapping is a strong default, not a rigid template — if the project lacks one of these elements (no stories, zero equations, etc.), drop or substitute a slide rather than showing a zero. See references/content-sourcing.md for the exact extraction commands for every slide.

Slide 12 exists to close the loop before the CTA: by slide 11 the reader has seen ten different facts in ten different visual styles — slide 12 pulls the single strongest claim out of each of slides 2-11 into one scannable list, so a reader who only skims gets the whole pitch in five seconds before hitting the ask on slide 13. Pick the best fact per slide, not every fact — this slide fails if it turns back into a wall of text.

Workflow

Step 1: Gather Source Material

Read, in parallel:

1. mkdocs.yml — site_name, site_description, site_url, repo_url, and theme.palette.primary / theme.palette.accent (drives every slide's color scheme)
2. docs/about.md — pull the "Why This Book" / "Why This Intelligent Textbook" section
3. docs/learning-graph/book-metrics.json — the canonical metrics object (same file linkedin-announcement-generator and readme-generator use — keep numbers consistent across all three artifacts)
4. docs/img/mascot/character-sheet.md — name, species, and the pose table
5. docs/learning-graph/color-config.json (if present) — taxonomy category colors, used as a small legend on the learning-graph slide
6. docs/license.md — license type and URL
7. mkdocs.yml nav — chapter list, to hand-pick 4-6 standout titles for slide 4

If book-metrics.json is missing or stale, regenerate it first (bk-generate-book-metrics) rather than estimating numbers.

Step 2: Capture the Two Required Screenshots

Learning graph (slide 3): Run bk-capture-screenshot against the graph-viewer MicroSim (commonly docs/sims/graph-viewer/) at a large window size so labels are legible when cropped:

bashCopy code
~/.local/bin/bk-capture-screenshot docs/sims/graph-viewer

The full-graph screenshot shows all concepts at once — too dense to read at slide/thumbnail size. Crop to a readable cluster of 15-25 nodes that shows concept labels and a spread of taxonomy colors, using scripts/crop-screenshot.py:

bashCopy code
python3 scripts/crop-screenshot.py docs/sims/graph-viewer/graph-viewer.png \
  linkedin/slides/learning-graph-zoom.png --box 300,200,1400,1000

Adjust the --box left,top,right,bottom coordinates by opening the full screenshot first — don't guess blindly. If no crop region looks clean (e.g. the graph is too sparse or labels overlap everywhere), fall back to asking the user to manually pan/zoom the live viewer in a browser and take their own screenshot.

Flagship MicroSim (slide 7): Ask the user which MicroSim best represents the book if it isn't obvious (prefer one with strong visuals — turtle graphics, a chart, a network — over a plain form-based one), then:

bashCopy code
~/.local/bin/bk-capture-screenshot docs/sims/<chosen-microsim>

Step 3: Confirm the Outline Before Generating

Before writing any slide code, show the user:

• The 13-slide plan with the specific numbers it will display (e.g. "38 chapters, 450 concepts, 31 MicroSims, 338 glossary terms")
• Which chapters you picked to highlight on slide 4
• Which MicroSim you picked for slide 7
• Which 5-6 facts you picked for the slide 12 benefits recap

This is a quick sanity check, not a full design review — the numbers are objective (pulled from book-metrics.json), but the chapter/MicroSim/recap picks are subjective and worth a 10-second confirmation before building 13 slides around them.

Step 4: Set Up the pptxgenjs Project

bashCopy code
mkdir -p linkedin
cd linkedin && npm install pptxgenjs

Use a square 10in x 10in slide (pptx.defineLayout({name:'SQUARE', width:10, height:10})) — LinkedIn document posts render best at 1:1 on mobile feeds. If the user prefers the taller 4:5 format some carousels use, use width:8, height:10 instead; confirm with the user only if they express a preference, otherwise default to square.

Step 5: Build Each Slide

Read references/slide-patterns.md for the nine reusable pptxgenjs slide patterns (title/ cover, mascot-aside, image-aside, big-numbers, icon-row, pose-grid, badge-callout, checklist recap, closing CTA) and references/content-sourcing.md for exactly which field feeds which slide. Map the 13 slides in the table above onto these patterns — several slides reuse the same pattern with different content (e.g. slides 2, 3, 9 are all variants of "image on one side, text on the other").

Design rules — LinkedIn carousels are read on a phone, not projected:

• No more than ~25 words of body text per slide (this is a stricter cap than a lecture deck)
• Minimum 18pt body text, 36pt+ for headlines — small text is illegible in the LinkedIn in-feed thumbnail
• One idea per slide, always
• Dark background (primary color) for slides 1 and 13 only — bookends the deck; slides 2-12 stay light/white for readability and print/PDF friendliness
• Use the project's actual theme.palette.primary / accent from mkdocs.yml, not generic blues

Step 6: Generate and Verify

1. Run the Node.js script to produce the .pptx
2. Confirm exactly 13 slides were produced
3. Open each slide (or export slide 1, 3, 6, 7, 12, and 13 as PNGs) and visually check that no text overflows its box and that images aren't stretched or pixelated
4. Report slide count and file size to the user

Step 7: Prepare for LinkedIn Upload

LinkedIn's document-post uploader (the icon in "Start a post" that looks like a page/document, distinct from the photo icon) accepts .pptx, .ppt, .pdf, .doc, and .docx directly — no PDF conversion is required. If the user wants a smaller file or to strip editable metadata before sharing, offer to convert with LibreOffice if it's installed:

bashCopy code
soffice --headless --convert-to pdf --outdir linkedin linkedin/<kebab-title>-carousel.pptx

Tell the user LinkedIn will also ask for a document title shown above the carousel in the feed — suggest a short, benefit-led title (e.g. "13 Things That Make This Python Textbook Different"), not the book's formal title.

Remind the user of the same rule from linkedin-announcement-generator: the post caption should not contain the external site link (LinkedIn suppresses reach on posts with links in the body) — but a URL printed as text on slide 13 of the document itself is fine, since it's not a clickable hyperlink in the post body. If they're also posting caption text, hand off to linkedin-announcement-generator and tell them to paste the site URL as the first comment, same as always.

Output

Copy code
{project-root}/linkedin/
├── generate.js                       # pptxgenjs generation script
├── <kebab-title>-carousel.pptx       # the deliverable
└── slides/
    ├── learning-graph-zoom.png       # cropped learning-graph screenshot
    └── microsim-preview.png          # flagship MicroSim screenshot

Keep this directory outside docs/ (sibling to it, like presentation/ in textbook-to-presentation-generator) so MkDocs never tries to build it into the site.

Quality Checklist

• Exactly 13 slides
• Every number on every slide traces back to book-metrics.json (no invented stats)
• Learning-graph image is a legible zoomed crop, not the full dense graph
• Mascot poses shown match the character sheet's actual pose set
• Colors match the project's mkdocs.yml theme palette
• No slide exceeds ~25 words of body text
• Body text is 18pt+ everywhere
• Slide 12's benefits recap pulls one strongest fact per prior slide, not every fact
• Slide 13 has a clear URL and call to action
• File opens cleanly in PowerPoint/Keynote/Google Slides with no broken image links

Troubleshooting

Issue: Learning-graph screenshot is blank or shows a loading spinner Solution: The graph-viewer MicroSim needs more render time — see microsim-utils's screen-capture reference for increasing the timeout.

Issue: Cropped learning-graph image still looks too dense to read Solution: Increase the screenshot's --window-size before cropping (more native pixels per node), or ask the user to manually zoom the live viewer and screenshot that.

Issue: book-metrics.json is missing Solution: Run bk-generate-book-metrics (via book-installer) before proceeding — don't estimate or recount content by hand.

Issue: Project has no mascot yet Solution: Point the user at book-installer's learning-mascot guide first; slides 2 and 6 depend on mascot art existing.

Related Skills

• linkedin-announcement-generator — writes the post caption text; often paired with this skill's carousel as the attached document
• textbook-to-presentation-generator — the long-form lecture-deck sibling; reuse its pptxgenjs setup patterns but not its 25-35 slide scope or speaker-notes step
• microsim-utils (screen-capture reference) — the bk-capture-screenshot tool this skill depends on for slides 3 and 7
• book-installer (learning-mascot, book-metrics guides) — creates the mascot art and metrics file this skill reads
• readme-generator — another consumer of the same book-metrics.json, useful for cross-checking that numbers stay consistent across all generated artifacts