import InfoBox from '../components/InfoBox.astro';

The repository had one domain glossary, a file named `CONTEXT.md`, and the agent documentation told skills to read it by that name. Once the development blog had a vocabulary of its own, a single file no longer described the repository. This change renamed the glossaries to `GLOSSARY.md`, added a `GLOSSARY-MAP.md` at the root to list them, and wrote the agent-skills setup into the repository guide, so a session starting cold can find both without reconstructing the layout from history.

## Two contexts and a map

The map is deliberately short. It lists two contexts, the game at the root and the blog under `blog/`, and states the one relationship between them: the blog reads the game's git history as raw material and never modifies the game's code. Writing that sentence down, rather than leaving it implied by the directory layout, is most of what the map adds.

The rename touched 29 files, most of them by a single changed line: references in the docs, the ADRs, the blog skill, and a few code comments. Only the agent setup document changed shape. It had described a single-context layout and now describes two.

## Recording the agent setup

The same change wrote three choices into the repository guide: GitHub Issues as the tracker, the five default triage labels, and the multi-context layout. Checking the labels against GitHub turned up one gap. The `needs-info` label did not exist, so it was created. The other four were already in place.

<InfoBox title="The steering check" variant="note">
A rename like this leaves stale references behind, and the repository's steering check fails when a path named in the agent guides no longer exists. The change ran that check with the others. One script in the devcontainer still mentioned the old file name, but it refers to a file in another skill, not to this repository's glossary, so it was left as it was.
</InfoBox>

## Where the earlier posts meet it

The skills this setup describes had already been gathered into one provider-neutral directory, as [one skill, three agents](/posts/one-skill-three-agents#the-directory-was-part-of-the-interface) narrates. The blog's vocabulary came from [building the blog machine](/posts/building-the-blog-machine#start-with-the-boundaries), where the domain model was drawn before the skill existed.

The glossaries are short enough to read in one sitting. The next change that adds a term to either of them will be the first real test of the map, and of whether a session can tell which context a new word belongs to.