Skip to content

claude-code-cli.md

Latest commit

 

History

History
166 lines (123 loc) · 6.57 KB

claude-code-cli.md

File metadata and controls

166 lines (123 loc) · 6.57 KB

Claude Code CLI — Primary Coding Agent 🚀

The Main Driver: Claude Code CLI has become my primary coding agent thanks to recent improvements in plan mode, multi-agent architecture, and global plan persistence.

Why Claude Code CLI is Now #1

Claude Code's recent releases (v2.0.28+) introduced features that fundamentally changed how I approach AI-assisted development. The combination of improved planning, multi-agent coordination, and efficient codebase exploration makes it superior for most everyday coding tasks.

Key Differentiators

Plan Mode Superpowers:

  • Plans are now saved globally and referenced across sessions
  • Dedicated Plan subagent builds more precise, actionable plans
  • Plans persist as markdown files for review, sharing, and iteration
  • Execute approved plans with confidence knowing the context is preserved

Multi-Agent Architecture:

  • Launch specialized subagents for different tasks (explore, plan, implement)
  • Subagents can be resumed, maintaining context across interactions
  • Dynamic model selection — use the right model for each subtask
  • Parallel agent execution for maximum efficiency

Explore Subagent:

  • Haiku-powered efficient codebase searching
  • Reduces context overhead by intelligently navigating large codebases
  • Find relevant files and patterns without loading everything into context
  • Perfect for understanding unfamiliar codebases quickly

Plugin & Extensibility:

  • Plugin system (v2.0.12+) for custom commands, agents, and hooks
  • MCP server marketplace integration
  • Team-shareable configurations and workflows
  • Extend functionality without leaving the CLI

Integration with Providers

Claude Code CLI works with cost-effective providers by overriding the Anthropic API endpoint at the system level.

Primary Setup (GLM Coding Plan):

GLM provides an OpenAI-compatible API that works with Claude Code CLI. See the official setup guide:

Secondary Setup (Synthetic.new):

Synthetic.new also provides Claude Code compatibility. See:

How it works: Both providers require overriding the ANTHROPIC_BASE_URL and ANTHROPIC_API_KEY environment variables at the system level. The documentation above walks through the exact setup steps.

Note: You'll need to reconfigure environment variables when switching between providers (GLM ↔ Synthetic).

When to Use Claude Code vs Droid

Scenario Use Claude Code Use Droid
Quick iterations & tight loops ✅ Best choice Overkill
Single-file changes & refactors ✅ Best choice Overkill
Multi-file features ✅ Best choice Good option
Exploring new codebases ✅ Explore subagent Manual
Large projects with formal specs Consider ✅ Best choice
Team coordination with phases Consider ✅ Best choice
Spec-driven enterprise work Consider ✅ Best choice

Rule of thumb: Start with Claude Code for most work. Switch to Droid when you need formal specification phases or structured team workflows.

Pro Tips for Maximum Productivity

1. Use Plan Mode for Complex Tasks

# Before implementing, let Claude plan it out
claude --plan "Add user authentication with OAuth2"
# Review the plan, approve, then execute

2. Leverage the Explore Subagent

Don't manually grep through codebases. Let the Explore subagent find relevant files:

claude "Where is the authentication middleware implemented?"
# Explore subagent efficiently searches without loading entire codebase

3. Save and Reference Plans

Plans persist globally — reference them in future sessions:

claude "Continue implementing the auth plan from yesterday"
# Claude references the saved plan automatically

4. Parallel Agent Execution

For independent tasks, launch multiple agents:

claude "Run tests AND check for linting issues"
# Independent tasks run in parallel for faster feedback

5. Configure for Your Workflow

Provider switching requires updating environment variables. A simple approach:

# In your .zshrc or .bashrc, create functions to switch providers:

# Switch to GLM (primary daily driver)
glm-mode() {
  export ANTHROPIC_BASE_URL="https://api.z.ai/v1"  # Check GLM docs for exact URL
  export ANTHROPIC_API_KEY="your-glm-api-key"
  echo "Switched to GLM"
}

# Switch to Synthetic (for MiniMax/Kimi)
synthetic-mode() {
  export ANTHROPIC_BASE_URL="https://api.synthetic.new/v1"  # Check Synthetic docs
  export ANTHROPIC_API_KEY="your-synthetic-api-key"
  echo "Switched to Synthetic"
}

Important: Check the official integration guides for exact URLs and configuration:

Recent Improvements Worth Noting

  • Opus 4.5 & Haiku 4.5 support — Latest models with automatic fallbacks
  • Sandbox mode — Experimental BashTool sandboxing for safer execution
  • Improved permission prompts — Clearer UX for tool approvals
  • Session resume — Pick up where you left off seamlessly
  • Branch filtering — Better navigation in git-heavy workflows

Installation & Setup

# Install via npm
npm install -g @anthropic-ai/claude-code

# Verify installation
claude --version

Provider Configuration:

To use with GLM or Synthetic instead of native Anthropic, configure environment variables as documented in:

The Bottom Line

Claude Code CLI is now my primary coding agent because it strikes the perfect balance:

  • Fast for quick iterations and small changes
  • Powerful with multi-agent architecture for complex work
  • Efficient with Explore subagent for codebase navigation
  • Flexible with BYOK provider support (GLM, Synthetic, etc.)
  • Persistent with global plan storage across sessions

For most vibecoding workflows, Claude Code CLI gets the job done faster and more efficiently than alternatives. Reserve Droid CLI for when you genuinely need formal specification phases and structured team coordination.

Related:

Back: Development Tools