ClaudeGhost

Overview

Headless Supervisor for the Anthropic Claude CLI - Run Claude Code autonomously while staying in control via Telegram. ClaudeGhost wraps the claude CLI in a supervised shell, automatically approving safe operations based on your configured autonomy level while escalating risky commands to your phone for approval via Telegram Bot API.

The Challenge

Running AI coding assistants like Claude CLI requires constant supervision and manual approval of every command, making it impractical for autonomous workflows. Users need to balance automation with safety, but existing solutions either require constant attention or run completely unsupervised with no safety controls.

The Solution

Built a headless supervisor that wraps the Claude CLI with intelligent risk classification and multi-level autonomy controls. The system categorizes commands as Read-Only, Write, Execute, or Critical, and automatically approves safe operations based on configurable AFK levels (1-5). Risky commands are escalated to Telegram for instant mobile approval, providing the perfect balance between automation and control. Features include budget limits, session loops, automatic changelog tracking, and a live dashboard.

Key Features

• ✓AFK Autonomy Levels (1-5): From paranoid to god mode
• ✓Telegram Bot API: Approve/block commands from your phone
• ✓Risk Classification: Read-Only, Write, Execute, Critical
• ✓Budget Control: Set spending limits with automatic pause
• ✓Session Loop: Run multiple tasks back-to-back
• ✓Changelog Tracking: Automatic session logs
• ✓Auto-Update Checker: Get notified of new versions
• ✓Live Dashboard: Rich terminal UI with real-time status
• ✓Cross-Platform: Windows, macOS, and Linux
• ✓Stall Detection: Alerts when CLI hangs