graphics-designer

The on-brand graphics factory. HTML → PNG via headless Chrome, with mandatory adherence to your brand spec. Produces single graphics or multi-slide carousels. Logs every output in the Notion Content DB and mirrors the file to Drive's 03 Marketing/60 Graphics/.

Read <YOUR_BRAND_DOC> if you have one — voice rules + skill checklist + format-specifications table dictate dimensions, validators, and the visual discipline that separates on-brand from AI-generated.

When to use

Trigger on:

• "Design a graphic for X"
• "Make a LinkedIn portrait / square / banner"
• "Create a carousel about Y" (multi-slide)
• "I need a hero image / OG image / share image"
• "Render the X graphic"
• User explicitly types /graphics-designer

Skip for:

• Animated graphics → /launch-video (synthetic) or /video-use (footage edit)
• Slide decks → your decks pipeline (existing presentations/ build)
• PDF lead magnets → /lead-magnet-creator

Inputs

The brief should include:

• Format + Channel (e.g. LinkedIn portrait post, X header, carousel)
• Headline (use the two-line hero pattern · Outbound stack. / Live in 3 weeks.)
• Kicker (mono micro-label · "WEBINAR · 12 May" or "/research")
• Body (cards, bullets, table rows, speaker tiles · whatever the template has)
• Accent target (where the one accent dot / cell / pill lands)
• Optional template hint (which existing template to branch from)
• Optional source (Notion row URL · for cluster traceability)

If anything is missing, ask in one batch before drafting.

Workflow

1. Pick the right template

If you have existing templates in <YOUR_GRAPHICS_TEMPLATES_PATH>, branch from the closest match. Otherwise start from a clean HTML scaffold and apply your brand tokens (<YOUR_BRAND_TOKENS_PATH>).

Common template archetypes worth keeping in your library:

• Webinar event announcement (host + speakers)
• Webinar event with partner branding
• Tool / stack diagrams with logos
• Anatomy of a thing (annotated diagram)
• Accelerator / partner announcement
• Personal LinkedIn banner (1584×396)
• Company LinkedIn banner (1128×191)

If the brief doesn't fit any of the above, branch from the most modular template you have, or from your live styleguide HTML page. Don't reinvent · compose.

2. Branch the template

cp <YOUR_GRAPHICS_TEMPLATES_PATH>/<closest-match>.html \
   <YOUR_GRAPHICS_TEMPLATES_PATH>/<new-name>.html

Naming: <channel>-<topic>-<dateOrId>.html. Examples:

• linkedin-outbound-stack-2026-05.html
• linkedin-carousel-icp-tactics-01.html (carousel slide 1)
• linkedin-carousel-icp-tactics-02.html (carousel slide 2)
• x-header-2026-05.html

For carousels: one HTML file per slide. Same dimensions (1080×1350 each). Numbered suffix.

3. Edit copy + cards

Each existing template marks per-graphic edits in CONFIG blocks at the top of the file. Stay inside those blocks. Don't touch the design system styles · they're inherited from your tokens (mirrored inline).

Things you usually edit:

• <title> (browser tab + future PNG metadata)
• Header brand chip (Live · Webinar or /research slug)
• Kicker line (mono · short)
• Hero h1 (two lines · second in <em> for muted half)
• Speaker / card tiles (duplicate or remove rows as needed)
• Footer (URL, slug, page indicator)

Things you almost never touch:

• Token values (color, type, radius, spacing)
• Component classes (.brand, .kicker, .tick, .chip, .matrix)
• Layout helpers (.two-col, .three-col, etc.)
• The brand mark SVG · always present in .brand (path: <YOUR_LOGO_PATH>)

4. Render

cd graphics
./render.sh <new-name> <width> <height>

Defaults: 1080 1350 (LinkedIn portrait). For other formats, see the Format specifications table in <YOUR_BRAND_DOC>.

The render script is a thin wrapper around headless Chrome:

# render.sh pattern
"/Applications/Google Chrome.app/Contents/MacOS/Google Chrome" \
  --headless --disable-gpu --no-sandbox \
  --window-size=$WIDTH,$HEIGHT \
  --screenshot=output/$NAME.png \
  --default-background-color=00000000 \
  "file://$PWD/templates/$NAME.html"
# then a second pass at 2× device-scale-factor for the @2x output

For carousels: render each slide with the same command:

./render.sh linkedin-carousel-icp-tactics-01 1080 1350
./render.sh linkedin-carousel-icp-tactics-02 1080 1350
# ... etc

Outputs land in graphics/output/:

• <name>.png (1×, the deliverable)
• <name>@2x.png (retina · for high-density displays)

5. Visual review

Open the rendered PNG. Walk the skill checklist:

Copy checks:

1. Em-dash scan · zero — in any visible copy
2. Banned-words scan · zero hits on the slop list
3. Negation-pivot scan · no "not just X, but Y" patterns
4. Numerals where adjectives would weaken

Visual checks: 5. Accent count · the one accent appears at most once 6. Font discipline · body in Helvetica Neue 400, mono only on labels 7. Wordmark slot · <YOUR_BRAND> wordmark + logo mark visible (header or footer) 8. Hero-headline pattern · two lines, second in muted via .hero em 9. Footer URL · <YOUR_DOMAIN> (or campaign URL) 10. Corner ticks · all four .tick squares present, 35px clear 11. Dimensions match the format spec exactly

If any check fails: fix the HTML, re-render, recheck. Don't ship a graphic that fails any of the 11.

6. Upload to Drive

# parent: 03 Marketing/60 Graphics
# folder id: <YOUR_DRIVE_GRAPHICS_FOLDER>

Use mcp__claude_ai_Google_Drive__create_file per output PNG. Title pattern: <channel>-<topic>-<YYYY-MM-DD>.png (or @2x variant).

For carousels, upload all slides as separate files into a subfolder named after the carousel topic, so they stay grouped:

60 Graphics/
└── 2026-05 ICP tactics carousel/
    ├── slide-01.png
    ├── slide-02.png
    └── ...

Capture each viewUrl from the Drive create response.

7. Log in Notion

Create a row in the Content DB (<YOUR_NOTION_CONTENT_DB_ID>) via mcp__claude_ai_Notion__notion-create-pages:

• Title: <topic> · LinkedIn (e.g. Outbound stack · LinkedIn)
• Status: Scripting (awaiting human review) or Scheduled if the user confirms it's ready
• Pillar: pick the multi-select that matches the brief
• Format: Image for single graphic · Carousel / PDF for carousel
• Channel: LinkedIn (or whatever the brief specified)
• Drive Assets: paste the Drive viewUrl(s). For carousels, paste all slide URLs.

Page body: include the brief, the headline, the kicker, the source URL if invoked from /repurpose, and any notes for the human reviewer.

If invoked from /repurpose, the orchestrator may handle the row creation · check the source context first to avoid duplication.

8. Report back

One-line summary plus links:

• Rendered PNG path(s)
• Drive viewUrl(s)
• Notion row URL
• Confirmation that all 11 skill-checklist items passed

Suggest next: post on LinkedIn (manual · <YOUR_NAME> handles publishing), or chain into /linkedin-copywriter for the post copy that will accompany the graphic.

Carousel mode

LinkedIn carousels are the highest-engagement format. Build them deliberately.

Structure (default 7 slides)

Adjust slide count for the topic. 5-12 slides is the sweet spot. Less than 5 is just a graphic; more than 12 loses people.

Visual continuity across slides

• Same headline color treatment on every slide.
• Slide number in the footer (02 / 07) as mono.
• Accent rotates · don't anchor it to the same position every slide.
• Kicker stays the same across slides (it's the carousel's identity).

Render carousel

for i in 01 02 03 04 05 06 07; do
  ./render.sh <carousel-name>-$i 1080 1350
done

Don'ts

• Don't reinvent the design system. Compose existing component classes. If you need a new component, lift it from another template or from your live styleguide HTML. Adding new tokens or components belongs in <YOUR_BRAND_TOKENS_PATH>, not in a one-off graphic.
• Don't mix print and motion subsystems. A graphic is print-style (corner ticks, dot grid, accent). Motion-system surfaces (glass, hero video, drift orbs) are for /launch-video and web.
• Don't ship without re-rendering. If you edited the HTML, re-run ./render.sh and re-check the PNG. Trust the render, not your expectation.
• Don't skip the @2× output. The render script produces both; upload both. LinkedIn auto-picks the right density.
• Don't use stock photos. Speaker / customer photos go in graphics/public/speakers/ and graphics/public/webinar/. If you don't have a real photo, use a <div class="photo placeholder"> with initials, not a generic image.
• Don't bypass the skill checklist. All 11 items, every time.
• Don't write the post copy here. Once the graphic is shipped, hand off to /linkedin-copywriter for the accompanying text.

See also

• Brand reference: <YOUR_BRAND_DOC> · format spec table, voice rules, skill checklist
• Existing templates: <YOUR_GRAPHICS_TEMPLATES_PATH>
• Render script: graphics/render.sh
• Tokens: <YOUR_BRAND_TOKENS_PATH>
• Notion Content DB: <YOUR_NOTION_CONTENT_DB_ID>
• Drive folder 60 Graphics: <YOUR_DRIVE_GRAPHICS_FOLDER>
• Downstream skills: /linkedin-copywriter (post copy after graphic ships), /repurpose (when graphic is a derivative of a master)