The cloud is a vocabulary, and this week it splits in two. One set of words is shared — “skill” (1,830), “agent” (1,364), “claude” (1,039) — the language a thousand different repositories reach for when they describe building and extending AI coding agents. The other set is single projects whose names grew big enough to read as topics: “odysseus” (1,020), “hermes” (666). The first group tells you what the week was about; the second tells you what it was naming things.
Read past the headline words and a quieter naming pattern surfaces — the “open-” prefix. “open” lands at 973, carried by open-notebook, open-design and open-code-review: repos that state their purpose in a single morpheme, an open version of something closed. When a prefix trends as its own word, it has become a genre.
Look at the project names together and a fashion appears: the agents are going mythological. “odysseus” (1,020) and “hermes” (666) — a wanderer and a messenger — are the week’s two biggest project-name tokens, with “astrid” trailing down the list. It is the obverse of the “open-” pattern: where one camp names a project by function (open-this, code-that), another reaches for a proper name with a story attached. Both are bets on how you make an agent memorable — describe it, or christen it.
What’s absent is as loud. No “llama,” no “gpt” — “llm” (416) is the closest, and it’s a category, not a product. The week’s vocabulary is about the layer around models — agents, skills, harnesses — not the models themselves. The words people pick to name projects after have moved one level up the stack.
A caveat worth keeping: this is tokenized from names, so it measures how projects are named, not a curated taxonomy, and a popular project inflates its own words. But that is exactly why it works as a leading indicator — people name things after the world they think they are building. This week, they think they are building agents, and naming them, more and more, in Claude’s dialect.