opc-carousel-creator

Create an on-brand Oak Park Construction Instagram post/carousel (before-after, project progress, product tip) from photos or a topic. Use whenever Priscila asks to make/create an OPC post, carousel,

SKILL.md

GitHub

OPC Carousel Creator

Trigger: "make/create a post/carousel/before-after" for Oak Park Construction, with photos or a topic. Never start designing from imagination. This skill exists because a session once invented a navy/orange "brand" — that must never happen again.

Reference-loading order (MANDATORY, in this order, before any design work)

  1. docs/OPC_DESIGN_SYSTEM.md in priihigashi/oak-park-ai-hub (fetch via ~/bin/gh api if repo not local) — tokens, fonts, layout rules, template list, folder routing
  2. scripts/content_creator/opc_template_catalog.json — pick the template by purpose/good_for (progress/proof → opc_progress.html; tips → opc_tip.html; never invent a layout)
  3. CONTENT_FORMATS Drive doc 1XqXSyJC_iHMTrmMxpM5ZR7S-WQxz19HhDJO1HomdncM — check if a registered FORMAT applies
  4. The target Drive folder — in the Marketing SHARED drive (0AIPzwsJD_qqzUk9PVA) → Content → (Proof Posts 1R4p51rUyGSfgf5VMgFKjQVXl5A399_QI / Carousel 16P2JN74JAAW3HKnmNqPGPrAq7N5jDNii / Manual Posts 1NoWWdL9s9mIoevloioCceUKFnW6ncRFa); never My Drive, always supportsAllDrives=true — prior builds for version numbering (vN_proof-<slug>)

Workflow

  1. Intake: photos (which is before/after?) + one-line description. Category: Talking Head (→ video pipeline, not this skill) / Progress-Before-After / Product Tip.
  2. Clean photos: PIL re-save strips EXIF/GPS (Image.open → exif_transpose → resize ≤1440w → save JPEG q90). Never send GPS-tagged originals anywhere.
  3. Adapt the real template (fetch docs/templates/opc_<x>.html + CSS + fonts from the repo): fill the slide spine with real copy — plain contractor voice, no promises/fake stats, NO invented names/addresses/dates/crew/project-IDs, hook <10 words. Fill or remove every image slot — no dashed placeholders in a real post.
  4. Review gate BEFORE rendering: run /opc-carousel-reviewer checklist against the HTML drafts.
  5. Render 3 variants (the template's own v1 dark / v2 cream / v3 lime-on-dark) via Playwright/headless Chromium at 1080×1350, screenshot each .slide element. NEVER text-to-image AI.
  6. Open renders for Priscila (open folder) → she picks a version → finalize caption (150-200 chars + hashtags ≤30).
  7. Archive to Drive (needs her permission-prompt approval for file transfer on this Mac): Proof Posts/vN_proof-<slug>/ with png/, html/, originals_used/ subfolders per convention. Update Library Index/Sheet row if a new folder was created.
  8. Deliver Buffer package: file list in slide order + approved caption. Never publish/schedule without her explicit approval.

Done check

  • 3 variants rendered from an approved template, reviewer checklist passed
  • She compared and picked; caption approved
  • Package delivered (local paths + caption); Drive archived (or her-action noted)
  • No step skipped silently — report anything blocked, honestly

Working reference implementation

First build (2026-07-16 addition post): build_post.py pattern — fetch template/CSS/fonts via gh api, string-fill v1 slide block, clone for v2/v3, playwright.sync_api element screenshots. Reuse it.