Output Styles

An output style decides the personality and structure of Klaus's answers. Switching styles only changes how it expresses itself — not what it can do or what it is allowed to do.

• The style list = the two built-in personalities + whatever you write yourself. The built-ins are only Explanatory (explains, as it works, why it chose this implementation and what the conventions in the repo are) and Learning (pauses so you can write a small piece of code yourself for practice). The other entry, "default", is not a style — it means "apply no style at all".
• To write your own: drop in a .md file, and the filename becomes the style id. Lookup order is the global ~/.klaus/output-styles first, then each .klaus/output-styles walking up from the session's working directory level by level (stopping at the repository boundary), where a later match with the same name overrides the earlier one — so a project can carry a style that only applies inside that project. In the front matter, omitting name falls back to the filename, and omitting description falls back to the first line of the body.
• The selected style is stored as an engine setting and sent down during session initialization. The engine ships no built-in styles of its own; it just picks one by name from the list it receives. A session that is already open is rebuilt automatically between turns with the new style, so you don't have to reopen it on purpose.
• It only rewrites the "how to express yourself" part of the system prompt. The tool list and permissions are untouched, so switching styles never makes Klaus more capable or more restricted. Only name, description and the body in the front matter are actually used. Fields such as keep-coding-instructions, which circulate online, can be written into the file, but the engine does not read them and they change no behavior — if you want a pure chat personality, just write the body that way.