AiCodeBoost - Auto-Approve for Claude Code & Codex

Stop clicking "approve" every time. Code 10x faster with AI.
The Problem
You're using Claude Code or OpenAI Codex CLI to supercharge your coding. But there's a catch:
- Click approve. Click approve. Click approve.
- Dozens of permission popups per session
- Your flow is constantly interrupted
- AI can't work autonomously
You didn't install an AI assistant to babysit it.
The Solution
AiCodeBoost configures smart auto-approval for your AI coding tools. One click, and your AI can actually help you code.
- Safe mode - Auto-approve reads only (zero risk)
- Balanced mode - Auto-approve file operations (daily coding)
- YOLO mode - Auto-approve everything (trusted projects)
All modes include built-in safety protections. Your .env files and dangerous commands are always blocked.
Why Developers Trust AiCodeBoost
|
|
| 100% Local |
No servers, no accounts, no data collection |
| Open Source |
Audit every line of code on GitHub |
| No Telemetry |
We literally can't track you |
| Safe by Default |
Dangerous operations always blocked |
| One-Click Setup |
Works instantly after install |
| Free Forever |
MIT licensed, no premium tiers |
Features
Three Approval Modes
| Mode |
What's Auto-Approved |
Best For |
| Safe |
File reads, searches, glob patterns |
Maximum security |
| Balanced |
+ File edits, writes |
Daily coding workflow |
| YOLO |
+ Bash commands, web requests |
Trusted projects |
Built-in Safety (Always Active)
Even in YOLO mode, AiCodeBoost always blocks:
rm -rf and destructive commands
curl | bash and remote execution
.env, .env.local, .env.production
secrets/, credentials, API keys
Works With
| Tool |
What It Configures |
| Claude Code (Anthropic) |
~/.claude/settings.json |
| Codex CLI (OpenAI) |
~/.codex/config.toml |
Easy Controls
- Sidebar Panel - Visual mode selector
- Status Bar - One-click mode cycling
- Command Palette - Full control
- Keyboard Shortcut - Instant switching
Phase Runner (NEW in v1.1)
Run complex AI tasks in sequential phases. Perfect for large projects that need to be broken down into manageable steps.
- Plan your project in phases using ChatGPT or Claude
- Paste all phases into the Phase Runner tab
- Click Start - AI executes Phase 1, then Phase 2, and so on
- Monitor progress with visual indicators and controls
Example workflow:
## Phase 1: Project Setup
Create the folder structure, package.json, and TypeScript config...
---
## Phase 2: Database Models
Set up Prisma schema with User and Product models...
---
## Phase 3: API Routes
Create REST endpoints for CRUD operations...
Pro tip: Ask ChatGPT or Claude to plan your project:
"Plan building [your app] in phases. Format each phase with ## Phase N: Title followed by detailed instructions. Separate phases with ---"
Installation
From VS Code
- Open Extensions (
Cmd+Shift+X / Ctrl+Shift+X)
- Search "AiCodeBoost"
- Click Install
- Done. It just works.
From Terminal
code --install-extension AiCodeBoost.aicodeboost
Usage
Change Mode (3 Ways)
1. Sidebar
Click the AiCodeBoost icon in the activity bar. Select your mode.
2. Status Bar
Click the mode indicator (bottom-right) to cycle modes.
3. Keyboard
- Mac:
Cmd+Ctrl+Shift+A
- Windows/Linux:
Ctrl+Shift+Alt+A
Commands
Open Command Palette (Cmd+Shift+P / Ctrl+Shift+P):
AiCodeBoost: Set Approval Mode
AiCodeBoost: Cycle Approval Mode
AiCodeBoost: Open Panel
AiCodeBoost: Start Phase Runner
AiCodeBoost: Stop Phase Runner
AiCodeBoost: Pause/Resume Phase Runner
AiCodeBoost: Skip to Next Phase
How It Works
AiCodeBoost writes configuration files that Claude Code and Codex read on startup:
Claude Code (~/.claude/settings.json):
{
"permissions": {
"defaultMode": "bypassPermissions",
"allow": ["Read(**)", "Glob(**)", "Grep(**)", "Edit(**)", "Bash(*)"],
"deny": ["Read(./.env*)", "Edit(./.env*)"]
}
}
Codex CLI (~/.codex/config.toml):
approval_policy = "on-request"
FAQ
Is this safe to use?
Yes. AiCodeBoost only modifies configuration files for Claude Code and Codex. It doesn't have access to your code, can't execute commands, and sends zero data anywhere. The blocked patterns (.env, rm -rf, etc.) are hardcoded and can't be bypassed.
Will this work with GitHub Copilot?
No. Copilot doesn't have an approval system - it's inline suggestions only. AiCodeBoost is specifically for Claude Code and Codex CLI which use permission dialogs.
What's the difference between modes?
- Safe: Only auto-approves reading files (searches, grep, glob). AI can explore but not change anything.
- Balanced: Also auto-approves writing/editing files. AI can make changes but can't run shell commands.
- YOLO: Auto-approves everything except dangerous operations. Use on projects you fully trust.
Can I customize blocked commands?
The safety blocks are intentionally hardcoded to prevent accidents. For advanced customization, you can manually edit the config files after AiCodeBoost creates them.
Does this send any data?
No. AiCodeBoost is 100% local. There's no server, no analytics, no telemetry. We couldn't track you even if we wanted to.
Is this free?
Yes, forever. MIT licensed, open source, no premium features.
What if I break something?
Each mode creates a standard config file. You can manually edit or delete ~/.claude/settings.json or ~/.codex/config.toml to reset.
Requirements
- VS Code 1.85.0+
- Claude Code and/or Codex CLI installed
Privacy & Security
- No data collection - Nothing leaves your machine
- No accounts - No signup required
- No telemetry - Zero tracking
- Open source - View the code
Support
License
MIT License - Use it however you want.
Created by Houston IT Developers
Making AI coding tools faster and safer for developers everywhere.