What the agent does
The agent starts with the layout chain. In a blog like 11ty/eleventy-base-blog, every page renders through _includes/layouts/base.njk, whose <head> carries a <title> of {{ title or metadata.title }}, one description meta tag with the same fallback, and the Atom feed link; there is no og: or twitter: tag anywhere. Posts use _includes/layouts/post.njk, which declares layout: layouts/base.njk in its front matter, so a post's title, description and date already reach the base layout. Site-wide values live in _data/metadata.js: title, url (still https://example.com/), language, description and an author object. eleventy.config.js registers HtmlBasePlugin, and content/sitemap.xml.njk already turns a relative page.url into an absolute one with page.url | htmlBaseUrl(metadata.url), which is the same filter the share tags need.
It then adds one block to the base layout's <head>. og:site_name and og:locale come from metadata.title and metadata.language; og:title and og:description reuse the fallbacks the <title> tag already uses; og:url is page.url | htmlBaseUrl(metadata.url). og:type is article when the page's tags include posts, which content/blog/blog.11tydata.js adds to every post, and website otherwise, and an article also gets article:published_time from page.date | htmlDateString, the filter _config/filters.js defines for the sitemap. The twitter:card tag is summary_large_image when an image is set and summary when none is.
Images need an absolute URL that the eleventy-img transform does not rewrite, because that plugin only processes image elements in the body, such as the ./possum.png in content/blog/fourthpost/fourthpost.md. So the agent adds an optional image front matter key, documented in content/blog/firstpost.md as an example, and a default share image path in _data/metadata.js; a file under public/img/ reaches the site root through the addPassthroughCopy({"./public/": "/"}) line in eleventy.config.js. It extends _data/eleventyDataSchema.js, where a Zod schema already validates the draft key, with image: z.optional(z.string()), so a typo in the key fails the build the same way an invalid draft does.
The proof is the built HTML. The agent runs npm install and npm run build, which is npx @11ty/eleventy on the Node 22 that .nvmrc pins, then reads _site/index.html, _site/blog/firstpost/index.html and _site/about/index.html to show the tags on a home page, a post and a plain page, and that og:url is absolute. It leaves metadata.url as it found it and says in the pull request that the share tags are only right once that value is the real domain, the same caveat the sitemap already carries.
Start it with one API call
POST https://api.phoenix.so/o/{org}/sessions with this body:
{
"agent": "claude",
"input": "Add Open Graph and Twitter card meta tags to this Eleventy site. Read eleventy.config.js, _data/metadata.js, _data/eleventyDataSchema.js, _includes/layouts/base.njk, _includes/layouts/post.njk, content/blog/blog.11tydata.js and two posts first. In the base layout add og:site_name, og:locale, og:title, og:description, og:url (absolute, through the htmlBaseUrl filter with metadata.url), og:type (article for pages tagged posts, website otherwise), article:published_time for articles, og:image and twitter:card, with twitter:card falling back to summary when there is no image. Support an optional image front matter key with a site-wide default in metadata.js, validate the key in eleventyDataSchema.js, add one example to a post, and keep every existing head tag. Run npm install and npm run build, show the tags in the built HTML for the home page, one post and the about page, and open a pull request that notes metadata.url must be set to the real domain.",
"resources": [
{
"url": "https://github.com/your-org/your-app"
}
]
}- Get an API key. In the console, open Settings, then API keys, and create a key for your org. Send it in the Authorization header as a Bearer token.
- Send the request. POST the body in request.json to https://api.phoenix.so/o/{org}/sessions, with your repository as the resources URL. The run starts in its own cloud sandbox.
- Read what came back. Follow the run on its live stream or wait for the run.completed webhook, then read the result and its trajectory.
What comes back
What comes back is a pull request on its own branch touching _includes/layouts/base.njk (the new meta block), _data/metadata.js (a default share image path), _data/eleventyDataSchema.js (the image rule), one post's front matter as the example, a share image under public/img/, and a short README note on the image key. eleventy.config.js, the post layout and the other posts are unchanged.
The description quotes the rendered <head> of a post and of the home page from _site, so a reviewer sees og:type switch between article and website and sees the absolute og:url without building anything. The build output is in the transcript at GET /o/{org}/sessions/{id}/messages beside the command that produced it.
Before merging, a reviewer sets url in _data/metadata.js if it is still the placeholder, opens one built page in a card validator, and decides whether the default image is the one they want. A follow-up message to the same session can add og:image:alt or an article:author tag from metadata.author.
Run it on every Linear issue labelled seo
A team that tracks site work in Linear starts one run per issue from an automation with a linear trigger on the seo label. The automation adds the issue after this prompt, so an issue that asks for a particular default image or for article:author as well is what the agent reads, and the run starts in the repository the automation names.
Each run opens its own pull request from its own sandbox, and an automation run has its own max_runs_per_hour cap. The pull request links the issue and quotes the built HTML, so the review is a read of one layout file and one rendered page.
The use case this job follows, with its automation and trigger: Turn a Linear label into a PR.
Why run it on Rush?
Share tags are a small change that is easy to get subtly wrong: a relative og:url, a twitter:card that names an image the page does not have, a key that posts spell three ways. The agent reads the real layout chain and the real filters, builds the site with the repository's own npm run build, and checks the output HTML rather than trusting the template, in a sandbox with the pinned Node and the lockfile's packages.
The pull request carries the built <head> for three kinds of page in its description and the full build log in the transcript. Several sites can be handled at once, each in its own sandbox, each with one pull request.
Frequently asked questions
Does this change how my posts are written?
Only by adding an optional image key. Posts without it keep working and fall back to the site-wide default in _data/metadata.js; title, description and date are read from the front matter the posts already have.
Why not use the eleventy-img plugin for og:image?
The transform plugin in eleventy.config.js rewrites image elements in the page body and writes hashed files, which is right for inline images but gives a meta tag nothing to point at. A plain path under public/, copied to the site root by the existing passthrough rule, gives a stable absolute URL. The pull request notes how to generate a sized image with the plugin's shortcode later if the team wants one per post.
What if metadata.url is still example.com?
The tags render with that domain, exactly as content/sitemap.xml.njk already does. The agent does not guess the real domain; it says in the pull request that url in _data/metadata.js must be set, and the reviewer sets it once for both.
Will the build fail on a bad image key?
Yes, on purpose. _data/eleventyDataSchema.js already runs a Zod schema over every page's data and throws on an invalid draft; the agent adds image to that schema so a number or a list in the key stops the build with a readable error instead of producing an empty tag.