viben executor

Discover and view executors. Executors are the coding agents that actually run Agents, such as Claude Code, Cursor, etc.

Concept Overview

What is an Executor?

An Executor is the underlying coding agent that runs an Agent. Viben assembles feature-rich Agents by configuring skills, prompts, MCP servers, etc. for these executors.

Concept	Description	Examples
Executor	Underlying coding agent responsible for task execution	Claude Code, Cursor, Gemini CLI, Codex, Windsurf
Agent	Viben-configured agent instance based on an executor	`main` agent (based on Claude Code)
Relationship	Agent = Executor + Skills + Prompts + MCP + Memory	One executor can support multiple agents

Executor Responsibilities:

Receive user instructions
Call LLM for reasoning
Execute tool calls (code writing, file operations, etc.)
Communicate with MCP servers

Viben Responsibilities:

Discover locally installed executors
Configure skills, prompts, MCP for executors
Manage agent instances (based on executors)
Provide unified configuration and memory management

Supported Executors

ID	Name	Description	Detection Method
`CLAUDE_CODE`	Claude Code	Anthropic official CLI	`claude --version`
`CURSOR`	Cursor	AI-first editor	`cursor --version`
`GEMINI_CLI`	Gemini CLI	Google Gemini CLI	`gemini --version`
`CODEX`	OpenAI Codex	OpenAI Codex CLI	`codex --version`
`WINDSURF`	Windsurf	Codeium IDE	`windsurf --version`
`AMP`	Amp	Sourcegraph Amp	`amp --version`
`OPENCODE`	OpenCode	Open source coding agent	`opencode --version`
`QWEN_CODE`	Qwen Code	Alibaba Qwen coding agent	`qwen-code --version`
`AIDER`	Aider	AI pair programming	`aider --version`
`CONTINUE`	Continue	IDE plugin	`continue --version`

Commands

# ============================================================
# Executor Discovery (Discovery Only)
# ============================================================

# List supported executor types
viben executor types
viben executor types --json

# List all discovered executors (with installation status)
viben executor list
viben executor list --json

# View specific executor details
viben executor show -n <executor-id>
viben executor show -n CLAUDE_CODE
viben executor show -n CURSOR --json

# ============================================================
# Executor Chat (Non-interactive execution)
# ============================================================

# Basic usage
viben executor chat -n CLAUDE_CODE -p "Analyze this code"

# Read from stdin
echo "Write a sorting function" | viben executor chat -n CLAUDE_CODE

# JSON stream input/output (for programmatic calls)
viben executor chat -n CLAUDE_CODE --input-format stream-json --output-format stream-json

# Resume session
viben executor chat -n CLAUDE_CODE -p "Continue" --resume <session-id>

备注

Discovery functionality does not implement installation; users should install via each executor's official channels.

Output Examples

viben executor types (Human-readable):

TYPE          DESCRIPTION
------------  -----------------------
CLAUDE_CODE   Claude Code (Anthropic)
AMP           Amp
GEMINI        Gemini CLI (Google)
CODEX         Codex CLI (OpenAI)
OPENCODE      Opencode
CURSOR_AGENT  Cursor Agent
QWEN_CODE     Qwen Code (Alibaba)
COPILOT       GitHub Copilot
DROID         Droid

viben executor types --json:

{
  "success": true,
  "data": {
    "types": [
      { "id": "CLAUDE_CODE", "name": "Claude Code (Anthropic)" },
      { "id": "AMP", "name": "Amp" },
      { "id": "GEMINI", "name": "Gemini CLI (Google)" },
      { "id": "CODEX", "name": "Codex CLI (OpenAI)" },
      { "id": "OPENCODE", "name": "Opencode" },
      { "id": "CURSOR_AGENT", "name": "Cursor Agent" },
      { "id": "QWEN_CODE", "name": "Qwen Code (Alibaba)" },
      { "id": "COPILOT", "name": "GitHub Copilot" },
      { "id": "DROID", "name": "Droid" }
    ]
  }
}

viben executor list (Human-readable):

Executors:

  Installed:
    CLAUDE_CODE     Claude Code      v1.0.25    Anthropic's official CLI
    CURSOR          Cursor           v0.45.2    AI-first code editor

  Not Installed:
    GEMINI_CLI      Gemini CLI       -          Google Gemini CLI
    CODEX           OpenAI Codex     -          OpenAI Codex CLI
    WINDSURF        Windsurf         -          Codeium IDE
    AMP             Amp              -          Sourcegraph Amp
    OPENCODE        OpenCode         -          Open source coding agent
    QWEN_CODE       Qwen Code        -          Alibaba Qwen coding agent
    AIDER           Aider            -          AI pair programming
    CONTINUE        Continue         -          IDE plugin for AI coding

Tip: Use 'viben executor show -n <id>' to see details.

viben executor show -n CLAUDE_CODE (Human-readable):

Executor: CLAUDE_CODE
Name: Claude Code
Description: Anthropic's official CLI for Claude

Status: ✓ Installed
Version: 1.0.25
Path: /usr/local/bin/claude

Configuration:
  Config Dir:    ~/.claude/
  MCP Config:    ~/.claude/mcp_servers.json
  Settings:      ~/.claude/settings.json

Agents using this executor:
  main          3 sessions    default
  my-agent      1 session

Capabilities:
  - Tool use (Bash, Read, Write, Edit, etc.)
  - MCP server support
  - Multi-turn conversations
  - Extended thinking
  - Image understanding

viben executor list --json:

{
  "success": true,
  "data": {
    "executors": [
      {
        "id": "CLAUDE_CODE",
        "name": "Claude Code",
        "description": "Anthropic's official CLI for Claude",
        "installed": true,
        "version": "1.0.25",
        "path": "/usr/local/bin/claude",
        "configDir": "~/.claude/",
        "mcpConfigPath": "~/.claude/mcp_servers.json",
        "settingsPath": "~/.claude/settings.json"
      },
      {
        "id": "CURSOR",
        "name": "Cursor",
        "description": "AI-first code editor",
        "installed": true,
        "version": "0.45.2",
        "path": "/Applications/Cursor.app/Contents/MacOS/Cursor",
        "configDir": "~/.cursor/",
        "mcpConfigPath": "~/.cursor/mcp.json",
        "settingsPath": "~/.cursor/settings.json"
      },
      {
        "id": "GEMINI_CLI",
        "name": "Gemini CLI",
        "description": "Google Gemini CLI",
        "installed": false
      }
    ]
  }
}

Detection Interface

Viben uses a typed detection interface to discover executors on the system. Each supported executor has a detector definition that specifies how to find and identify it:

interface ExecutorDetector {
  id: string;                    // e.g., "CLAUDE_CODE"
  name: string;                  // e.g., "Claude Code"
  description: string;           // e.g., "Anthropic's official CLI for Claude"
  detectCommand: string;         // e.g., "claude --version"
  configPaths: string[];         // e.g., ["~/.claude/", "~/.config/claude/"]
  mcpConfigPath?: string;        // e.g., "~/.claude/mcp_servers.json"
  settingsPath?: string;         // e.g., "~/.claude/settings.json"
}

The detect() method runs the detectCommand for each executor and returns a DetectedExecutor result:

interface DetectedExecutor {
  id: string;
  name: string;
  description: string;
  installed: boolean;
  version?: string;              // Parsed from detectCommand output
  path?: string;                 // Executable file path
  configDir?: string;            // Configuration directory
  mcpConfigPath?: string;        // MCP configuration file path
  settingsPath?: string;         // Settings file path
}

You can use the --timeout option with the detect/list commands to control how long Viben waits for each executor's detection command to respond:

# List executors with a custom timeout (in milliseconds)
viben executor list --timeout 5000

Executor Configuration

Executor detection can be configured via YAML. Each executor definition includes detection commands used to identify whether the executor is installed:

# Example executor config
id: CLAUDE_CODE
name: "Claude Code"
description: "Anthropic's official CLI for Claude"
detection_commands:
  - command: "claude --version"
    parse_version: true
  - command: "which claude"
    parse_path: true
config_paths:
  - "~/.claude/"
  - "~/.config/claude/"
mcp_config_path: "~/.claude/mcp_servers.json"
settings_path: "~/.claude/settings.json"

The detection_commands section allows multiple commands to be run during detection. The parse_version flag indicates the command output should be parsed for a version string, while parse_path extracts the executable path.

Executor Chat

Non-interactive conversation with an executor.

Command Interface

viben executor chat [OPTIONS] -n <EXECUTOR_NAME>

OPTIONS:
    -n, --name <NAME>           Executor name (e.g., CLAUDE_CODE, GEMINI)
    -p, --prompt <PROMPT>       Prompt (optional, reads from stdin if not provided)
    -C, --cwd <DIR>             Working directory (default: current directory)

    --input-format <FORMAT>     Input format: text (default), stream-json
    --output-format <FORMAT>    Output format: text (default), stream-json
    --verbose                   Verbose output

    --session-id <ID>           Specify session ID
    --resume <SESSION_ID>       Resume existing session

    --model <MODEL>             Specify model (if executor supports)
    --dangerously-skip-permissions  Skip permission checks

Usage Examples

# Basic usage
viben executor chat -n CLAUDE_CODE -p "Analyze this code"

# Read plain text from stdin
echo "Write a sorting function" | viben executor chat -n CLAUDE_CODE

# JSON stream input/output (for programmatic calls)
echo '{"type":"user","message":{"role":"user","content":"Analyze code"}}' | \
  viben executor chat -n CLAUDE_CODE --input-format stream-json --output-format stream-json

# Resume session
viben executor chat -n CLAUDE_CODE -p "Continue previous work" --resume abc123

Relationship with Agents

┌─────────────────────────────────────────────────────────────┐
│                    Viben Architecture                       │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  viben executor list          viben agent list              │
│       ↓                            ↓                        │
│  ┌─────────────┐             ┌─────────────┐                │
│  │  Executor   │─────────────│   Agent     │                │
│  │ CLAUDE_CODE │   1:N       │   main      │                │
│  │ (Installed) │             │  my-agent   │                │
│  └─────────────┘             └─────────────┘                │
│       │                            │                        │
│       │ Runtime env                │ Configuration          │
│       ↓                            ↓                        │
│  ┌─────────────┐             ┌─────────────┐                │
│  │  ~/.claude/ │             │ ~/.viben/   │                │
│  │ (executor   │◄────────────│ agents/xxx/ │                │
│  │  native cfg)│  overlay    │ (skills,mcp │                │
│  └─────────────┘             │  memory)    │                │
│                              └─────────────┘                │
│                                                             │
│  Workflow:                                                  │
│  1. viben executor list  → Discover locally installed       │
│  2. viben agent create -n xxx -e CLAUDE_CODE                │
│     → Create agent instance based on Claude Code            │
│  3. viben agent config → Configure skills, mcp, prompts     │
│  4. When running agent, executor config + viben config merge│
│                                                             │
└─────────────────────────────────────────────────────────────┘

Key Differences:

Dimension	Executor	Agent
Definition	Underlying coding agent tool	Viben-configured agent instance
Quantity	Few (typically 1-3 installed)	Many (can create any number)
Management	System-level install, Viben only discovers	Viben creates and manages
Configuration	Native config (e.g., ~/.claude/)	Viben config (overlays native)
Operations	`list`, `show` (read-only)	`create`, `remove`, `config` (read-write)

executor chat vs agent chat

Feature	`viben executor chat`	`viben agent chat`
Target	Directly call underlying executor	Call configured Agent
Memory	None	Automatically loads Agent memory
Configuration	Command line args only	Agent config + workspace + command line
Session	Temporary	Persisted to Agent directory
MCP	Manual configuration needed	Uses Agent's MCP configuration
Skills	None	Uses Agent's Skills
Use Case	One-off calls, script integration	Ongoing conversations, project development

viben agent - Agent management
viben agent chat - Agent chat

Concept Overview​

What is an Executor?​

Supported Executors​

Commands​

Output Examples​

Detection Interface​

Executor Configuration​

Executor Chat​

Command Interface​

Usage Examples​

Relationship with Agents​

executor chat vs agent chat​

Related Commands​