claude-code-modes CLI
The claude-mode CLI launches Claude Code with a swapped behavioral prompt layer: presets from cautious safe to bold create
- 119 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The tool replaces part of Claude Code's system prompt, and presets like create give the agent more autonomy
- The install downloads a binary and runs it through the shell
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/claude-code-modesDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The tool replaces part of Claude Code's system prompt, and presets like create give the agent more autonomy
- The install downloads a binary and runs it through the shell
This is third-party code. Review the repository files before installing.
What it does
claude-mode launches Claude Code with its own system prompt. The default prompt makes Claude cautious, minimal and terse even where you want boldness and thoroughness. The tool keeps everything Claude Code needs to work: tool instructions, security and environment detection. Only the behavioral layer is replaced: how much initiative Claude takes, what code quality it targets and how far beyond the request it goes. Three axes set behavior: agency, quality and scope. Presets bundle them: `create` builds from scratch with proper architecture, `safe` makes surgical changes to production code, `refactor` restructures freely, `explore` only reads and explains, `debug` hunts the cause from evidence and `director` hands work to sub-agents. Other presets change character: `flow`, `tinker`, `spark`, `muse` and `straight` for candid judgment without cushioning, while `none` strips all instructions and leaves your CLAUDE.md. The prompt base is picked from the session model and can be swapped by hand: `lean`, `standard`, `chill`, `flow`, `straight`. The binary installs with one command or builds from source with Bun.
Who it is for. Developers unhappy with stock Claude Code's caution who want to pick a working style per task, from a production fix to a prototype.
Good fit when
- A new module is built from scratch and needs proper architecture: `claude-mode create`
- Surgical edits go into production code and Claude must stay in scope: `claude-mode safe`
- You need to understand unfamiliar code without a single file change: `claude-mode explore`
- You want a counterpart who directly challenges bad premises: `claude-mode straight`
Not a fit when
- You need a clean run with no system prompt changes: the behavioral layer is always replaced except in `none`
- You want to tune named sub-agents such as Explore and Plan: they do not see your prompt
- The run uses print mode: there is no git status in the prompt
Example request
claude-mode explore: explain how the auth module is built without changing any files.Limitations
Claude Code on the PATH is required, and Bun when installing from source; the binary needs no Bun. Named sub-agents such as Explore and Plan run on their own prompts and do not see your axes, while general-purpose sub-agents inherit your prompt. Print mode has no git status: since Claude Code 2.1.286 the working directory, platform and git arrive as separate messages. Self-update through `claude-mode update` works only on binaries from the upstream repository. The `curl | sh` install downloads a file and verifies its checksum.
How to disable. Delete the ~/.local/bin/claude-mode binary or remove the link with bun unlink if you installed from source. Delete the claude-mode config file by hand and then start Claude Code as usual with claude.
FAQ
How do I install claude-mode?
Run `curl -fsSL https://raw.githubusercontent.com/nklisch/claude-code-modes/main/install.sh | sh`: the binary lands in `~/.local/bin/claude-mode` with a SHA-256 check. A source build needs Bun, `bun install` and `bun link`.
Which preset suits production code?
`claude-mode safe`: collaborative agency, minimal quality and narrow scope. It makes surgical changes and stays on the request. For strict instruction following there is `methodical`.
How do I update it?
Run `claude-mode update`, `--check` to test without installing, or `claude-mode update 0.2.5` to pin a version. From a repository clone use `git pull && bun install`.
What do the lean, standard and chill bases mean?
A base is the system prompt template. `lean` and `standard` mirror what Claude Code sends the model, while `chill` is about a third shorter with calmer framing and no caps. The default depends on the model, and `--base` picks one by hand.
Do presets affect sub-agents?
General-purpose sub-agents inherit your prompt and axes. Named specialists such as Explore and Plan have their own prompts and run independently of the preset. Custom agents use whatever their own file says.
Discussion
No questions or reviews yet.
Related
CLI toolsClaude Code Blueprint: 12 agents, 19 skills, 16 hooks and 6 CLAUDE.md rules, copy what you need or install a preset
A handbook and starter set for Claude Code: guides, CLAUDE.md templates, starter kits, hooks and the commit-helper plugin.
A Claude Code environment installer: CLAUDE.md, rules, skills, agents, hooks and two-layer project memory in one script
Anthropic DevRel playground: three Claude Code mods, hooks, CLAUDE.md examples and skills, each with notes on how it was built.
A starter kit for Claude Code: a CLAUDE.md memory-bank template with ready hooks, subagents and skills you copy into your project
An installer giving Claude Code rules for phases, decision logs, debugging and tests in Go, Rust, Python and six more languages.
README in short
The README explains how it replaces Claude Code's system prompt and removes caution where it gets in the way. It covers installation, updates, the preset list with an axis table, prompt bases, configuration through a file, an approach to bold framing, limitations and sub-agent behavior. Separate sections cover the muse, flow and playful modes.