social-post-carousel
Rebuild a text post or user-supplied social screenshot as a clean, unbranded 1080 × 1080 carousel with exact copy, optional attribution, and numbered PNGs. Use for X-style social carousels, not for publishing or accessing accounts.
Try your first task
A source-grounded starter task has not been verified for this skill yet.
SKILL.md
name: social-post-carouseldescription: Rebuild a text post or user-supplied social screenshot as a clean, unbranded 1080 × 1080 carousel with exact copy, optional attribution, and numbered PNGs. Use for X-style social carousels, not for publishing or accessing accounts.Social Post Carousel
Repository integration
In a full checkout, use the repository's non-blocking version check when network access is allowed. Follow existing telemetry consent; never enable reporting here. If permitted, log only the skill name, elapsed time, outcome and version through the repository telemetry helper. Never log copy, filenames or avatars. Standalone copies may skip unavailable helpers.
Turn the source into legible, brand-neutral post cards. Re-typeset the text; do not crop whole screenshots into slides or ask an image model to draw the copy.
- Establish the source. Use supplied full text over screenshot OCR. If only an image is supplied, transcribe what is visible and flag clipped or uncertain text; do not reconstruct missing content. Default to verbatim copy, including order, figures and punctuation. Only edit the voice, add hooks or add a CTA when requested.
- Create the slide plan. Start with the original first line, set in the same style as all other copy. Break at paragraphs or idea boundaries, keeping comparisons and lists together when practical. Aim for four comfortably filled cards for a medium-length post. Combine short paragraphs before adding slides; avoid cards with unusually large blank areas. Add cards for long copy rather than shrink or truncate. Preserve handles in the copy as plain text, not blue links, by default. Use a name/handle and avatar only if supplied and appropriate; no verification badge, platform logo, engagement counters, status bar or navigation. Do not fabricate an avatar.
- Render. Use a solid black background, off-white text, gray attribution
and one fixed body size from first card to last. Place the optional author and handle on one line. Do not add display headlines, divider rules, or page counters inside the images by default; the output filenames carry slide order. Keep at least 78 px safe margins.
Write a JSON spec with a
slidesarray (strings with paragraph breaks) and optionalname,handleandavatar(path relative to the JSON). Use the deterministic Pillow helper with regular and bold sans-serif fonts already installed on the machine. Do not commit screenshots, input JSON, logos or personal images to a public skill; the included example is synthetic. - QA and deliver. Review a contact sheet at phone scale and every PNG at full size when needed. Verify 1080 × 1080 dimensions, slide order, text coverage and absence of UI, clipping or accidental blank cards. If a slide overflows, split or rebalance it; do not shrink individual slides. Compare the contact sheet against any supplied visual reference before delivery. Deliver ordered PNGs (and a ZIP or PDF only as helpful). Do not post to a social account unless separately authorized.
Run
Requires Python 3.10+ and Pillow. From the skill folder:
python scripts/render_carousel.py examples/demo.json out \
--regular /path/to/regular.ttf --bold /path/to/bold.ttf
The renderer is offline; it never fetches external images or publishes. For languages outside the font's coverage, use a font with the necessary glyphs and inspect the output. For complex script shaping or vertical typography, use a capable alternative renderer while retaining the same layout checks.