A GitHub Repo as a Database

npx skills@latest add phucbm/skills --skill github-as-db

clyrics is a library of Chinese song lyrics with pinyin and translation, for people learning the language. It keeps the lyrics as JSON files in its own repo. A visitor adds a song in the app; a GitHub App bot makes a branch, commits the file and opens a pull request, and the song is live once I merge it. Reads never touch the API: a script compiles every file into public/data/index.json before each build.

Why a repo

The pull request is the moderation step, git history is the audit log, and issues are the support queue. I pay for none of them. It only works because the data is public, small and written rarely; anything private, real-time or relational needs a real database.

What broke

• btoa breaks on Chinese. The API wants file content in base64, and btoa only takes Latin-1, so the first song with a Chinese title threw. btoa(unescape(encodeURIComponent(json))) fixes it.
• Reading each file through the API burns the rate limit. 60 requests an hour without a token, one per song. The build-time index turns it into one static file.
• Every update needs the file's current SHA. Without it GitHub answers 422, so an update or a delete reads the file first.
• The private key in the browser. clyrics is a Vite app, so the App's key ships in the bundle. I accepted it because the repo is public, the App can only touch that one repo, and the worst it can do is open a pull request I still have to merge. In Next.js the same flow goes in a server action instead.