Command Code

The first frontier coding agent that both builds software and continuously learns your coding taste.

Ships full-stack projects, features, fixes bugs, writes tests, and refactors — all while learning how you write code.

---

Install

npm i -g command-code

Run

cd your-project && cmd

Explore the quickstart guide →

---

Take command of your code

Models Opus, Sonnet, Haiku Claude frontier LLMs	Taste taste-1 meta neuro-symbolics Learns how you code	Extend CLI, Skills, MCP Build on top of it
Code 10x faster Ship features fast	Review 2x quicker Slash review time in half	Bugs 5x slashed Catch issues early

---

What you can do

• Build projects in anyone's taste — match any coding style automatically
• Automate the tedious stuff — repetitive tasks, boilerplate, scaffolding
• Review PRs in seconds — get instant, contextual feedback
• Create commits and pull requests — ship directly from the CLI
• Refactor without breaking things — safe, taste-aware transformations
• Plan before you build — structured reasoning before code generation
• Connect your tools with MCP — extend with Model Context Protocol servers
• Customize with skills and taste — tailor behavior to your workflow
• Write tests and docs — generate comprehensive coverage and documentation
• Pipe, script, and automate — composable CLI for any workflow

Explore common workflows and examples →

---

How taste works

Continuously Learning Every accept, reject, and edit becomes a signal that shapes your taste profile.

Meta Neuro-Symbolic AI taste-1 enforces the invisible logic of your choices and coding taste.

Share with your team Portable via npx taste push/pull. Rules decay. Taste compounds.

Explore how taste learns your style →

---

Explore the docs

Topic	Description
Core Concepts	Interactive, headless, and plan modes.
Taste	Learns your coding style and applies it to every project automatically.
Skills	Reusable slash commands to share.
MCP	Connect external tools and services via Model Context Protocol.
Studio	Manage taste profiles, API keys, and team organizations.
Troubleshooting	Common issues, FAQs, and errors.

---

Quickstart

1. Install

npm i -g command-code

2. Navigate to your project

cd your-project

3. Launch

cmd

On first launch, you'll be guided through authentication and initial taste calibration.

---

Core Concepts

Command Code operates in three modes:

Mode	Description	Usage
Interactive	Conversational coding with real-time feedback	cmd (default)
Headless	Non-interactive execution for CI/CD and scripts	cmd --headless "task"
Plan	Structured planning before code generation	cmd --plan

Models

Command Code is powered by Claude frontier LLMs:

• Opus — Maximum capability for complex, multi-step tasks
• Sonnet — Balanced performance for everyday development
• Haiku — Fast responses for quick edits and simple tasks

Switch models with the /model command during an interactive session.

---

Taste

taste-1 is a meta neuro-symbolic system that continuously learns your coding preferences.

How it works

1. Observation — Every accept, reject, and edit you make is recorded as a signal
2. Modeling — taste-1 builds a symbolic representation of your coding patterns
3. Enforcement — Your taste profile is applied to all generated code automatically

Sharing taste profiles

# Push your taste profile to the team
npx taste push

# Pull a team member's taste profile
npx taste pull

Rules decay. Taste compounds.

---

Skills

Skills are reusable slash commands that extend Command Code's capabilities.

# List available skills
/skills

# Use a skill
/skill:deploy

Create custom skills as markdown files and share them with your team.

---

MCP

Command Code supports the Model Context Protocol for connecting external tools and services.

Built-in MCP servers

Command Code ships with GitHub's MCP server by default.

Custom MCP servers

Configure additional MCP servers in your project's .command/mcp.json:

{
  "servers": {
    "my-server": {
      "command": "npx",
      "args": ["my-mcp-server"]
    }
  }
}

---

Studio

Manage your Command Code environment:

• Taste profiles — Create, edit, and switch between taste profiles
• API keys — Configure authentication and model access
• Team organizations — Manage team members and shared configurations

---

Common Workflows

Build a new feature

cmd "add dark mode toggle to settings page"

Fix a bug

cmd "fix the race condition in the auth flow"

Review a PR

cmd "review PR #42"

Write tests

cmd "write tests for the payment module"

Refactor

cmd "refactor the database layer to use connection pooling"

Pipe and automate

cat issue.md | cmd --headless

---

CLI Reference

devin entry point

devin                            # Start interactive REPL (no prompt)
devin -- your prompt here        # Start REPL with initial prompt
devin -p "prompt"                # Single-turn, no REPL: print response to stdout and exit
devin -p -- prompt words here    # Same, using -- separator (still works)

Flag	Description
-p, --print	Single-turn mode — print response and exit
-m, --model <name>	Select model (opus, sonnet, haiku)
--headless	Non-interactive mode for CI/CD and scripts
--plan	Structured planning before code generation
-h, --help	Show help message
-v, --version	Show version

cmd entry point

Usage: cmd [options] [prompt]

Options:
  --headless          Run in non-interactive mode
  --plan              Enter plan mode
  --banner            Show the splash banner
  --experimental      Enable experimental features
  --model <name>      Set the model (opus, sonnet, haiku)

Slash Commands:
  /model              Switch the active model
  /skills             List available skills
  /skill:<name>       Run a specific skill
  /taste              View or edit your taste profile
  /login              Authenticate with your account
  /feedback           Submit feedback
  /experimental       Toggle experimental features
  /lsp                Check configured LSP servers

---

Troubleshooting

Issue	Solution
command not found: cmd	Ensure command-code is installed globally: npm i -g command-code
Authentication failed	Run /login to re-authenticate
Model not available	Check your subscription plan supports the selected model
MCP server not connecting	Verify your .command/mcp.json configuration
Taste not applying	Run npx taste pull to sync your latest profile

---

Contributing

Contributions are welcome! Please follow these guidelines:

1. Fork the repository
2. Create a new branch

   git checkout -b feature/your-feature-name

3. Make your changes
4. Test your changes
5. Commit and push

   git commit -m "Add feature: your-feature-name"
   git push origin feature/your-feature-name

6. Open a pull request

Issues & Suggestions: Open an issue for bugs, questions, or feature requests using the provided issue templates.

Feedback from the CLI: Run /feedback from within the CLI to submit a confidential feedback survey.

---

Next Steps

• Quickstart — Walk through your first real task
• Common Workflows — Real examples of what to build
• CLI Reference — Every command, flag, and configuration option
• Core Concepts — Interactive, headless, and plan modes

Explore the community on Discord →

---

License

This project is licensed under the Command Code License.

---

Contact

• Organization: ATC-O48
• Project Link: https://github.com/ATC-O48/copilot-cli

---