Generate OG and social images for every page in one prompt

Give every page of my site a generated social/OG image, hosted on PixelVault.

Context - read this first and use only the documented API:
- Overview + API: https://pixelvault.dev/llms.txt (sections "Generate an image from a prompt" and "Image Transforms")

Do this:
1. Find where my site builds each page's <head> (a layout or SEO component) and list the pages that have no og:image today.

2. Create a script (for example scripts/generate-og.mjs, with an npm script "generate-og") that, for each page missing one:
   - builds a prompt from the page title and description plus ONE shared style sentence in a single constant, and ALWAYS includes "no text, no letters, no logos" (the image models cannot render text reliably);
   - calls POST https://api.pixelvault.dev/v1/images/generate;
   - stores the returned data.url in a manifest file (for example og-images.json, keyed by route) so each page is generated once.

3. In the head layout, when the manifest has a URL for the current page, add these tags using the URL plus ?size=social, which PixelVault serves as a 1200x630 social card:
   <meta property="og:image" content="URL?size=social">
   <meta property="og:image:width" content="1200">
   <meta property="og:image:height" content="630">
   <meta name="twitter:card" content="summary_large_image">
   <meta name="twitter:image" content="URL?size=social">
   Pages without an entry keep my existing default image.

4. Run it on at most 3 pages first and show me the URLs, then tell me how to check one with a social-card validator.

Result I want: each page has a distinct 1200x630 social image on a permanent img.pixelvault.dev URL, generated once and stored in the manifest.

Rules:
- Call the API from server-side or build-time code only. The key is a SECRET key (pv_live_...): read it from the PIXELVAULT_API_KEY environment variable, never hard-code it, never put it in browser code or in a public env var (no NEXT_PUBLIC_, PUBLIC_ or VITE_ prefix), and make sure .env is listed in .gitignore.
- Use only the documented endpoint: POST https://api.pixelvault.dev/v1/images/generate with header "Authorization: Bearer $PIXELVAULT_API_KEY" and JSON body {"prompt": "...", "model": "flux-1-schnell"}. The hosted image is in data.url (data.optimized_url is a smaller WebP/AVIF). Do NOT invent options, endpoints or packages.
- Each image costs 1 credit on the default model and the free plan has 10 credits a month. Before generating a batch, call GET https://api.pixelvault.dev/v1/usage and stop if data.generation.credits_remaining is lower than the number of images you plan to make. Never generate in an unbounded loop: cap the number of images per run.
- Errors: 402 (out of credits, or a model my plan does not include) means stop and tell me, do not retry. 429 means wait for the Retry-After header. 502, 503 and 504 are safe to retry (credits are refunded), at most twice. 422 prompt_rejected means change the prompt.
- Generated images are not repeatable. Never regenerate an image that already has a saved URL unless I ask.
- Keep the change minimal and match my existing code style.
- Do not put text in the generated image. If I want the page title on the card, tell me instead of trying to draw it in the prompt.