One file, two readers

This site already serves two audiences. Browsers get the HTML you are looking at. Agents get llms.txt, a persona skill, and an MCP server. The blog follows the same rule.

Every post here is one markdown file in the repository, with a short frontmatter block for the title, dates, and tags. The build renders that file to HTML for this page. An agent can request the same file directly:

curl -H "Accept: text/markdown" https://frankponcelet.com/blog/one-file-two-readers

Appending .md to the URL works too, for fetchers that cannot set headers. Both return the raw source of this post, byte for byte, frontmatter included. The RSS feed is generated from the same files.

Why bother? Because a second copy is a second thing to maintain, and copies drift. I would rather keep one file and let the server decide how to present it. The frontmatter parser is about forty lines and has no dependencies. The markdown rendering uses marked, which has none either.

More posts are coming, starting with the story of the CryptoPunks V1 wrapper and the design decisions behind ERC721F.