What are output styles in AI coding tools?
By Flavio
Output styles guide a coding agent's tone and response format. Learn how they differ from verbosity, reasoning summaries, and project instructions.
Output styles are instructions that shape how a coding agent responds, including its tone, length, and format. Claude Code uses this name for a session-wide feature.
How to make AI harnesses talk less covers the practical settings. Start there if you want shorter replies.
How output styles work
The agent’s application adds style instructions to the model’s context. Those instructions influence the words it generates. The application separately decides whether to display tool logs or reasoning summaries.
This is also different from reasoning effort, which controls a model’s reasoning budget or behavior.
In coding tools
Claude Code provides styles including Concise, Explanatory, and Learning. It sends the active style with each request. Custom styles can also change Claude’s role.
Codex’s configuration reference lists separate controls for model verbosity and reasoning summaries. Support depends on the selected model. The older personality values friendly and pragmatic no longer select a style.
Cursor User Rules support communication preferences across projects. A rule asking for short replies is a prompt instruction.
Grok Build’s command reference documents /compact-mode for a denser interface. A session-wide output-style selector is not documented on that page.
Details checked against each vendor’s documentation in November 2026.
A concrete example
To select Claude Code’s Concise style for one project, merge this field into .claude/settings.local.json:
{
"outputStyle": "Concise"
}
Keep any existing settings. The name is case-sensitive, and the new style applies from your next message. For a detailed code explanation, ask for the detail you need in that message.
When to use output styles
Use a persistent style when you want the same response format throughout a session. Use a one-off prompt when only one answer needs a different shape.
Pitfalls
Short replies still need enough information to explain failures. Keep test failures and decisions requiring your attention visible.
A prose instruction cannot hide interface logs. Check the application’s display controls when the extra output comes from tools.
When writing a custom Claude Code style, check keep-coding-instructions. It preserves software engineering instructions in the full system prompt; the shorter prompt omits that section.
Related on flaviocopes.com
- How to make AI harnesses talk less separates response instructions from interface controls.
- AGENTS.md explains project instructions shared with coding agents.
- Discover CLI tools for coding agents shows why useful command output still matters when replies are short.
Related glossary entries
- System prompt: A system prompt is the instruction layer a coding agent sends ahead of your chat. How Claude Code, Codex, Cursor and Grok Build set that layer.
- Rules: Project rules are durable instructions a coding agent loads for a repo. How they work in Cursor, Claude Code, Codex and Grok Build, and when to skip them.
- Reasoning effort: Reasoning effort sets how long a model thinks before it answers. The levels in Claude Code and Codex, what Cursor's plans fix, and how Grok sets effort.
Want me to talk about your product? You can sponsor this site.