Author Once, Compile Everywhere
Author Once, Compile Everywhere
Every AI assistant has invented its own file format for standing instructions: Kiro has steering files, Claude Code has skills, Codex has its own, Copilot has instructions files, Cursor and Windsurf and Cline each have rules. Maintaining the same guidance in eight formats is the problem Kanon exists to solve.
You author one knowledge artifact — a metadata block plus a written body — and the pipeline turns it into each tool's native output, degrading gracefully where a tool lacks a feature. You can see the whole idea by comparing one artifact's source with its compiled form, which is what this exercise does; running the compiler yourself is optional.
The kanon skill's authoring guide walks this in nine steps, and it teaches artifact metadata using Dublin Core — because it was written for library staff. Use it rather than this page; reading it as instructional design is half the exercise.
For a library, the interesting part is not the plumbing. It is that professional practice becomes a versioned, describable, reviewable object that outlives whichever product your institution is subscribed to this year. That is the same argument you already make about metadata standards.
Open the authoring guide
Do not work from this page. The kanon skill carries a nine-step authoring guide written for library staff — use it, and read it as a piece of instructional design while you do.
Use the `kanon` skill and give me its authoring guide for creating a first knowledge artifact. Summarise the nine steps, then tell me what its metadata-quality checklist requires and what its most common listed mistakes are.
You have nine steps from idea to compiled output, plus a metadata checklist — and you notice its worked examples are Dublin Core.