LinkedIn Carousel

A document post is one idea broken into steps the reader swipes through. It fails when a single claim is cut into slides. The user uploads the PDF; this skill only writes and builds it.

Phase 1: Is it a carousel?

1. Read the Voice section of 01-Perfiles/Perfil Maestro.md (/master-profile), as /linkedin-post does. No voice yet: stop and send the user there.
2. The idea needs sequence: steps, a progression, a framework with parts, a before and after. One claim, or fewer than four parts, is a text post: say so and hand over to /linkedin-post.
3. A technical topic follows /linkedin-post Explain mode: one concept, one reader, every claim checked against the primary source before any slide is written.

Phase 2: Slides

8-12 slides is the working range: fewer is usually a text post, and past 12 cut a step rather than add a slide.

1        COVER     the hook in 6 words or fewer, plus one line of promise
2        STAKE     why this matters to the reader, one sentence
3..N     ONE IDEA PER SLIDE: a 3-7 word headline, at most 25 words under it
N+1      RECAP     the whole thing as a list, readable without the deck
LAST     CLOSE     one action: follow, comment, or the link in the first comment

• The cover decides the result. Pick it from three options drawn from .claude/skills/linkedin-post/references/hooks.json, and say which you would ship.
• A slide that needs a paragraph is two slides.
• Number every slide (3/10) and put the user's name or handle small in a corner: screenshots travel without the post.
• Code on a slide: six lines at most, monospace, the one line that matters highlighted.
• Never fabricate numbers, results or experience; a missing number is {{...}} and a question.

Phase 3: Post text and humanizing

The carousel still needs 2-3 lines above it, and those lines are the hook the feed shows. Write them, then run the post text and the slide copy through /content-humanizer:

python .claude/skills/content-humanizer/scripts/humanize.py copy.txt -o clean.txt --report
python .claude/skills/content-humanizer/scripts/detect.py copy.txt clean.txt

Show the slides as a numbered list the user can read in ten seconds, plus the post text. Build the PDF only after the user approves the copy.

Phase 4: The PDF

Write slides.html with one <section> per slide and print it:

<style>
  @page { size: 1080px 1350px; margin: 0 }
  body { margin: 0 }
  section { width: 1080px; height: 1350px; box-sizing: border-box;
            padding: 96px; page-break-after: always; overflow: hidden }
</style>

python .claude/skills/linkedin-carousel/scripts/carousel_pdf.py slides.html carousel.pdf

• 4:5 (1080x1350) is the common choice because it takes more of a phone screen than a square. That is practice, not a LinkedIn rule; LinkedIn only requires every page of a PDF to be the same size, which the @page rule guarantees.
• Type no smaller than 28px: most people read it on a phone.
• Colors and fonts: use the project's brand skill or design system if there is one. Otherwise one accent color on a neutral background, and no invented palette.
• The script prints with headless Chrome or Edge and checks LinkedIn's limits (100 MB, 300 pages). It also checks that there is one page per slide: a [WARN] about extra pages means a slide overflows and shows up cut in the feed. Shorten that slide and print again.

Phase 5: Hand over

CAROUSEL READY
cover:     #19 Curiosity Gap
slides:    10 (PDF 10 pages, 1080x1350)
humanizer: human score 81 PASS
title:     "How the KV cache works, in 10 slides"   (LinkedIn asks for one on upload)
file:      carousel.pdf
link:      goes in the first comment

Reply "yes" to log it, or tell me what to change.

On "yes", ask "May I append it to 01-Perfiles/LinkedIn Log.md?" and log the date, the cover and the format.

Protocol

• Never upload, publish or schedule, and never open LinkedIn in a browser. The User Agreement (section 8.2) forbids automated posting.
• Credit: adapted from Jake Schincariol's linkedin-agent-skill (MIT).

Next steps

• /linkedin-comment for the engagement round before posting.
• /linkedin-reply once comments arrive.