Why Claude Code is the best fit
Claude Code can read files, run shell commands, and edit your repo in-place. That maps 1:1 to the workflow we care about: open your README, pick a few cards from our 21-endpoint API, write the image tags, save. With /llms.txt as context, Claude doesn't have to guess parameter names — it knows them, including how to URL-encode spaces and special chars.
The minimum prompt
Open Claude Code in your username/username repo (the special profile repo GitHub uses to render a README at the top of your profile). Then type something like: "Read https://coolreadme.xyz/llms.txt and build me a README using my GitHub username yerdaulet-damir. Pick 4 cards that fit a backend engineer who ships every day. Write the result to README.md." That is enough. Claude will fetch the docs, pick cards (cinematic, hacker, cat-card, ai-card is a common combo), construct the URLs, and write the file.
What Claude knows from /llms.txt
Every endpoint, every parameter, encoding rules, decision trees for "what card fits this user", and example scenarios for a backend engineer / OSS maintainer / streamer / etc. We deliberately optimized the file to be skimmed by an LLM in one pass — no marketing fluff, all schema and examples.
Animated cards only
Every card we generate is a live API endpoint — either a Satori-rendered PNG or raw SVG with embedded CSS animations. There are no static images to download or assets to host. When GitHub renders your README, it hits our edge-cached endpoint and the card updates if you change a query param. Pet streak cards even re-fetch your GitHub contribution graph and evolve through stages.
Tips for working with Claude Code
Tell Claude to verify each URL renders before committing — it can curl them with `-I` and check for `Content-Type: image/svg+xml` or `image/png`. Ask it to add `width="100%"` on `<img>` tags so cards stretch on mobile. If your README looks crowded, ask Claude to add a horizontal `---` between sections.