arch-doc: reference architectures as concept explainers, always credited

A demo proves that one thing works. A reference architecture explains an idea well enough that you could build your own: What is an agent harness? What is AI Ops? How does an agent remember? arch-doc writes those explainers in one house format.

Ask Claude for "reference architecture: agent harness" or "explain CoALA for my site". The skill:

• frames the concept in one sentence;
• finds the sources that named it;
• teaches it in seven short sections with two or three hand-drawn diagrams;
• links the demos on this site that implement it.

Concepts here, code in the demos

An architecture never links a repository. It explains the idea and points to the demos that implement it under See it built. Each of those demos carries a clickable Follows tag back to the concept. You can start from either end: read the idea, then see it running, or find a demo and learn the idea behind it.

Credit is part of the format

Every explainer ends with a numbered Credits & sources list. Each entry names the paper, the vendor's documentation, the talk or the project, and says what was taken from it. The skill verifies every URL it cites and never invents an author or a year. Diagrams adapted from a source are redrawn and captioned "adapted from …". On this site the rule is enforced: an architecture without sources fails the build.

“Inspired by best practices is not a credit. Name the paper, the repo or the talk, and what you took from it.”

See it in use

The first explainer on this site, CoALA: how language agents remember, decide and act, was written with arch-doc. It maps the RAG demo onto the framework.