agent-cli

cck-agent

A mini coding assistant built with claude-code-kit -- a tiny Claude Code in ~120 lines.

What it does

• Reads, searches, and edits files using real tools from @claude-code-kit/tools
• Auto-detects API keys from environment variables (Anthropic, OpenAI, DeepSeek, SiliconFlow, Groq, Ollama)
• Falls back to a realistic mock demo when no API key is found
• Read-only tools (Glob, Grep, Read) auto-approve; destructive tools (Bash, Edit, Write) prompt for permission

Run

pnpm install
pnpm --filter agent-cli-example start

Connect a real LLM

Set any supported provider's API key:

# Anthropic (default)
export ANTHROPIC_API_KEY=sk-ant-...

# OpenAI
export OPENAI_API_KEY=sk-...

# DeepSeek
export DEEPSEEK_API_KEY=sk-...

# SiliconFlow
export SILICONFLOW_API_KEY=sk-...

# Groq
export GROQ_API_KEY=gsk_...

# Ollama (no key needed, just run ollama locally)

The CLI checks providers in order and uses the first one with a valid env var. No key found = demo mode with mock responses.

Tools

Tool	Permission	Description
Glob	Auto-approve	Find files by pattern
Grep	Auto-approve	Search file contents
Read	Auto-approve	Read file contents
Bash	Ask user	Run shell commands
Edit	Ask user	Edit existing files
Write	Ask user	Write new files

How it works

createAuth() -> try env vars -> fall back to MockProvider
                    |
                    v
          Agent({ provider, tools, permissionHandler })
                    |
                    v
              AgentREPL (interactive TUI)

All the auth, tools, permission, and UI come from claude-code-kit packages. The example just wires them together.