██████╗██╗      █████╗ ██╗   ██╗██████╗ ███████╗
██╔════╝██║     ██╔══██╗██║   ██║██╔══██╗██╔════╝
██║     ██║     ███████║██║   ██║██║  ██║█████╗
██║     ██║     ██╔══██║██║   ██║██║  ██║██╔══╝
╚██████╗███████╗██║  ██║╚██████╔╝██████╔╝███████╗
 ╚═════╝╚══════╝╚═╝  ╚═╝ ╚═════╝ ╚═════╝ ╚══════╝
████████╗██╗   ██╗██████╗ ██████╗  ██████╗
╚══██╔══╝██║   ██║██╔══██╗██╔══██╗██╔═══██╗
   ██║   ██║   ██║██████╔╝██████╔╝██║   ██║
   ██║   ██║   ██║██╔══██╗██╔══██╗██║   ██║
   ██║   ╚██████╔╝██║  ██║██████╔╝╚██████╔╝
   ╚═╝    ╚═════╝ ╚═╝  ╚═╝╚═════╝  ╚═════╝

Stop babysitting. Start building.

claude-turbo

Terminal wrapper that auto-accepts Claude Code permission prompts.

You know the drill. Claude wants to run a command. You press Enter. Claude wants to edit a file. You press 1. Claude wants to use a tool. You press Y. Over and over. Death by a thousand prompts.

claude-turbo wraps Claude Code in a PTY, watches for permission prompts, and auto-sends the right response. You see everything Claude sees — it just doesn't stop to ask anymore.

---

Usage

# Instead of:
claude

# Run:
claude-turbo

That's it. Everything else is transparent.

# With args
claude-turbo -- -p "fix the auth bug"

# Dry run — see what would be auto-accepted
claude-turbo --dry

# Custom delay before accepting (default: 300ms)
claude-turbo --delay 500

# Log every auto-accept to stderr
claude-turbo --log

What It Auto-Accepts

Pattern	Response	Example
"Proceed? (y/n)"	y	Build confirmation
"Allow / Deny"	1 (Allow)	Tool permissions
"Press Enter to continue"	Enter	Pause prompts
"Allow once / Allow always"	1 or 3	Permission scope
"Yes / No"	y	Generic confirmations
"Do you want to allow"	y	Tool use approval

What It NEVER Auto-Accepts

Safety patterns that always require manual input:

• delete all / destroy database / drop production
• force push to main/master
• rm -rf /
• this cannot be undone / are you sure / irreversible

If a dangerous pattern is detected, claude-turbo pauses and lets you decide.

How It Works

1. Spawns claude in a pseudo-terminal (PTY via node-pty)
2. All output flows through to your terminal — fully transparent
3. A rolling buffer watches the last 2000 chars for permission patterns
4. When a pattern matches, waits --delay ms, then sends the response
5. If you type anything during the delay, the auto-accept is cancelled
6. Dangerous patterns are blocked from auto-accept entirely

Install

git clone https://github.com/JacobiusMakes/claude-turbo.git
cd claude-turbo
npm install
npm link  # makes `claude-turbo` available globally

Config via Environment

CLAUDE_TURBO_DELAY=500   # ms before auto-accepting (default: 300)
CLAUDE_TURBO_LOG=1       # log auto-accepts to stderr
CLAUDE_PATH=/custom/claude  # custom claude binary path

Safety

• Dangerous patterns are blocked — delete/destroy/force-push always require manual input
• 300ms delay — gives you time to see what's happening and type to cancel
• User input cancels — if you type during the delay window, auto-accept is aborted
• Fully transparent — you see 100% of Claude's output, nothing is hidden
• Your existing permissions still apply — this doesn't bypass Claude's safety classifier, it just answers the prompts faster

License

MIT