What is seo-agent-pack?

seo-agent-pack is the actual working setup one person uses to write SEO articles for their own websites automatically, every night, using a normal Claude subscription instead of paying per API call. It is shared here as real files with two months of corrections already built in, not a polished demo made for GitHub. Each night a scheduled run picks one keyword from real search data, checks it against the pages currently ranking on Google for that term, writes an article following a strict set of rules meant to avoid generic AI content, generates matching images, and saves everything as a draft for a person to review. On Sundays it also rereads Google Search Console data and rewrites pages that are underperforming. Over two months it produced 70 articles and 65 pages across four sites, with every single piece checked by a human before it goes live. The project spends real effort explaining why this setup avoids becoming a spam machine. Keywords only come from verified search data tools, never invented. Before writing anything, the system studies the top three ranking pages and has to justify, in writing, why its own version deserves to beat them, or it skips the topic. Every draft passes through a pattern check and a separate pass built around a public list of common signs of AI writing. Facts, prices, and dates get checked against a live source during the run. Nothing publishes automatically, everything waits for a human to approve it. Setting it up takes about an hour and does not require coding. You install Claude Code, sign up for a keyword-data account, connect Google Search Console, get an image API key, and set up a Telegram bot for morning reports. Then you paste one included setup prompt into Claude Code, and it interviews you about your site, builds its own helper scripts, and asks for your keys locally. After a first test article, you review it, adjust anything you dislike, and turn on the nightly schedule. The author is upfront that first drafts still need tuning to a site's own voice, and that a brand-new site with no search history gets less benefit at the start. It is released under the MIT license.