Sai Poster Studio

Converts a text idea into a finished, on-brand poster for Sai (the product) /
Simular (the company). You (Claude) are the brain — you turn the user's text
into a structured brief; the local app renders it in the browser and exports.
There are no API keys: the LLM work is this Claude session, the rendering is
100% client-side.

Invocation

/sai-poster-studio <idea or post text>
/sai-poster-studio path/to/post.md

If invoked with nothing, ask the user for the idea / text in one line.

What it makes — templates (11)

Each template has a native size and brand-correct layout. You usually let the
brief auto-pick; override with template: in front-matter.

Template id	Use it for
cal-hacks (default)	events — date + venue, green "sai" tile + white card
hackathon-dark	hackathons — big title, date/venue/sponsors
happy-hour	mixers / drinks / socials
launch	launches, announcements, campaign statements (eyebrow / title / subhead / CTA pill / footer)
benchmark-dark	results — up to 3 big stats (e.g. OSWorld %)
talk-dark	talks / webinars / panels — speaker list + session code
editorial-cream	editorial features / quotes (Adamina serif, rotated sidebar)
feature-grid	2×2 feature cards
agent-hour	square two-tone with the brand butterfly
ascii-hero	full-bleed ASCII butterfly hero
enterprise	clean white + blue accent (healthcare / enterprise)

Sizes (set with the size picker in-app, or it follows the template):
instagram-square 1080×1080 · instagram-portrait 1080×1350 · linkedin 1200×627 · x 1600×900.

The brief (what you write)

A brief is Markdown. The H1 is the title; Key: value lines fill fields; a key
followed by a bullet list joins the bullets; front-matter picks type/template.

---
type: launch          # optional: event | launch | article — nudges the template
template: launch      # optional: force a template, skip auto-pick
---
# Don't wait for the future. Build it.   ← the H1 is the Title

Eyebrow: Sai 2.0
Subtitle: The agent that uses your computer for you.
CTA: Try Sai
Footer: simular.ai

Convenience fields (mapped for you): headline (or the # H1) → Title ·
proofLine → Proof/Subtitle · date · location · eyebrow · type · template.

Direct slot fields — use the exact label as the key:
Title, Title2, Subtitle, Eyebrow, Partners, Location, Date,
Category, Speakers, Featuring, Quote, Quote sub, Prize, Code,
CTA, Stat 1–Stat 3, Card 1–Card 4 (+ Card N body), Sidebar, Footer.
(\n in a value forces a line break.)

Auto-template priority: explicit template → explicit type
(event→cal-hacks/hackathon/happy-hour, launch→launch, article→editorial) →
keywords (hack→hackathon, mixer/happy hour→happy-hour, %/OSWorld/
benchmark→benchmark, Speakers→talk, Quote→editorial, date/location→
event) → cal-hacks.

Brand rules (non-negotiable)

• No emoji anywhere — strip them out of the user's text.
• Accent color is the signal-green #16D342 only. No other accent colors.
• Type is Manrope (headlines/body) + Adamina (serif accents, editorial).
• Keep copy tight: one idea per poster, short punchy headline, ≤ ~14-word subhead.
• Footer is usually simular.ai (or a campaign tag like #yourtag · simular.ai).

Workflow

Step 1 — Get the text

Use inline text if given; read the file if a path; if it's long, confirm it's complete.

Step 2 — Plan the brief

Pull out: the headline (the single sharpest line), an optional eyebrow
(series/context), a subtitle (one supporting line), a CTA, and a
footer. Pick (or let auto-pick choose) a template and set type:. Strip
emoji. Show the user the brief you intend to use before generating.

Step 3 — Generate (push it into the live studio)

cd ~/.claude/skills/sai-poster-studio          # the skill directory (this app)
[ -d node_modules ] || npm install             # first run only
# start the dev server in the background if it isn't already on :5181
(curl -sf localhost:5181 >/dev/null) || (npm run dev >/tmp/sai-studio.log 2>&1 &)
# write the brief to a temp file and push it onto the canvas
cat > /tmp/sai-brief.md <<'BRIEF'
---
type: launch
---
# Don't wait for the future. Build it.
Eyebrow: Sai 2.0
Subtitle: The agent that uses your computer for you.
CTA: Try Sai
Footer: simular.ai
BRIEF
npm run push /tmp/sai-brief.md                 # appears in the open editor within ~2s

npm run push is pure Node (no browser, no keys) — it parses the brief,
auto-picks a template, fills it, and writes the live document (.studio/current.json),
which the open editor loads automatically.

Step 4 — Preview & export

Tell the user to open http://localhost:5181. There they can:

• Edit text live in the Content panel; drag / resize / rotate elements; use the Align to canvas buttons and snap-to-element guides.
• Restyle: Background pattern (grid / dots / glow), font, color, highlight words, per-layer opacity.
• Add a background photo: select/add an image layer → Choose image (upload). (Briefs can't carry images — this is a manual step in the editor.)
• Switch size (Instagram / LinkedIn / X) — the layout reflows.
• Export PNG (2×/3×/4×, e.g. 3240²), SVG (vector), or JSON (re-editable).

To make more posters in the same session, just write another brief and
npm run push it again — the canvas updates live.

Editing capabilities (mention to the user as needed)

Free-form drag/resize, rotation handle (snaps to 15°), align-to-canvas + snap-to-element
guides, font picker (Manrope/Inter/Adamina/Instrument/Mono), per-word keyword highlighting,
decorative background patterns, per-layer opacity, layers panel, zoom, localStorage + disk sync.

Batch — dump many ideas, get all posters at once

When the user gives a list of ideas (e.g. "make 10 posters: …" or 10 lines/paragraphs), don't
push them one at a time. Write one brief per idea into a dedicated folder, render them all, and open
the gallery (a contact sheet of every poster). Needs npx playwright install chromium once.

mkdir -p briefs/<batch>                 # e.g. briefs/launch-week
# write one brief per idea, numbered so the gallery is ordered:
#   briefs/<batch>/01-slug.md, briefs/<batch>/02-slug.md, …
npm run generate briefs/<batch> --out posters/<batch> --scale 3
open posters/<batch>/gallery.html       # every poster in a grid; click any to open full size

gallery.html is a self-contained contact sheet written next to the PNGs — this is how the user
"views all of them at once." Rules: one idea per brief, number the files (01-…) for order, keep each
on-brand (strip emoji, tight copy), let templates auto-pick unless a type:/template: is obvious.
Any single poster can then be opened in the editor for fine-tuning: npm run push briefs/<batch>/03-foo.md.

Also available: npm run watch — auto-renders anything dropped into briefs/.

Common mistakes

Mistake	Fix
Emoji left in the copy	Strip all emoji — brand rule
Two ideas on one poster	One idea per poster; make a second
Headline too long	Tighten to a punchy line (display titles auto-shrink, but shorter reads better)
Wrong template for the content	Set type: or template: explicitly instead of fighting the keywords
Using a non-green accent	Accent is #16D342 only
Expecting a background image from the brief	Images are added in the editor (upload), not via the brief

Notes

• No API keys, ever. Claude writes the brief; the app renders in the browser. The only network use is loading Google Fonts.
• This skill bundles the full Vite app (src/, scripts/, briefs/). Don't edit the engine — just write briefs and push them.
• Source / updates: github.com/wyuee912-ops/sai-poster-studio