Claude Code Output Style Generator

This generator writes a Claude Code output style: a markdown file with YAML frontmatter that changes how Claude talks for the whole session, not just what it knows about your project. Six questions, all clicks except one optional line of free text, and what comes back is one file: a name and description for the /config picker, the one frontmatter flag people get wrong, and an instructions body built from the choices you made.

That flag is `keep-coding-instructions`, and it is the whole reason this file trips people up when written by hand. It defaults to false, which does not just change Claude's tone - it strips every one of Claude Code's built-in engineering habits, how it scopes changes, verifies its own work, writes comments. Set it true and you get a tone change on top of normal coding behavior; leave it false and you get a genuinely different assistant, which is correct for a writing or research persona and wrong for almost everything else. This generator asks the question directly instead of burying it in a table you'd only find by reading the docs.

Four presets cover the personas people actually ask for: a patient teacher that explains its reasoning as it goes, a terse operator that skips the preamble, a blunt reviewer that states problems without softening them, and a non-coding assistant for when Claude is drafting or analyzing rather than writing code. "Something else" drops the free-text box straight into the instructions body instead, so a persona that doesn't fit the four presets still gets a working file.

Output styles are one of several ways to shape how Claude Code behaves, and they're easy to confuse with the others. CLAUDE.md adds a message about your project and conventions; output styles rewrite the system prompt itself and apply to every response until you switch back. Neither one is enforcement - for a rule that must never break, like blocking a command outright, Claude Code hooks are the tool built for that job. Switching styles happens through `/config`, not the old `/output-style` command (removed in Claude Code v2.1.91), and a change only takes effect after `/clear` or a new session, never mid-conversation.

Pick personal scope and the file targets `~/.claude/output-styles/`, available in every project on your machine and never shared. Pick project scope and it targets `.claude/output-styles/`, meant to be committed so your whole team switches into the same voice. Once the style is set, the CLAUDE.md generator is the other half of a tuned setup - the file that tells Claude what your project actually looks like, versus this one, which tells it how to talk about it. For rules that need to hold regardless of which style is active, the Claude Code permissions guide covers the enforcement layer this file can't provide.

I run Claude Code daily and this generator writes the same frontmatter conventions my own output styles use. If you want the finished version of the whole idea - a tuned CLAUDE.md plus the full vetted tool stack, installed in one paste - that is what ClockedCode is. The link is at the bottom of this page.

By Neo Zino - builder of ClockedCode