Tired of shipping buggy plans? This AI-powered system ruthlessly critiques and perfects your implementation plans until they're bulletproof. No more "it should work" - only "it WILL work."
This tool is built specifically for users of the Codex platform utilizing the latest high-reasoning models.
To let the script "self-improve" and loop automatically until your plan is perfect, you MUST have:
- The Codex CLI installed:
npm install -g @openai/codex - Active API Credits on your OpenAI/Codex account.
- The Codex VS Code Extension installed and authenticated.
If you don't have the CLI or run out of credits, the script still works, but it is manual. You will have to copy-paste prompts into the VS Code extension for every single iteration until the judge is satisfied.
Imagine having a brutally honest AI reviewer that never lets you get away with:
- Hand-wavy assumptions
- Hidden edge cases
- Logical loopholes
- Vague requirements
This tool doesn't collaborate - it destroys weak reasoning and rebuilds your plans from the ground up, iterating until perfection. It leverages the specialized reasoning of the gpt-5.3-codex engine to enforce a strict zero-defect standard.
Just run the notebook! No command line required for basic setup.
- Open the setup notebook in VS Code or Jupyter:
- Run all cells - it will verify your environment and set up the foundation.
- Enable "Auto Mode": Run
npm install -g @openai/codexin a terminal. - Start using: Open a new terminal and type
plan-loop your-plan.md
cd automation_macOS
./INSTALL.sh
source ~/.zshrc
plan-loop your-plan.mdcd automation_windows
.\INSTALL.bat
# Restart terminal/IDE to refresh PATH
plan-loop your-plan.md- Feed it your draft - Rough ideas, incomplete thoughts, or full specs.
- Unleash the critic - The script parses your markdown and feeds it to the Codex 5.3 engine.
- The Loop (Fully Automated) - If you have the Codex CLI, the script automatically loops! It gets feedback, rewrites the plan, and keeps going until the judge gives a "True" status (perfection).
- The Manual Fallback - If the CLI is missing, the script generates prompt files (
prompt_judge.txt,prompt_rewrite.txt) for manual copy-pasting. - Claim victory - A perfect plan emerges, battle-tested and ready for execution.
- Viciously Critical Auditor: No mercy, no compromises.
- GPT-5.3-Codex Powered: Maximum reasoning power via
codex exec. - 999-Round Endurance: Stops only at perfection (strict pass:
pass=trueandproblems=[]). - State Preservation: Resume interrupted sessions safely via session markers.
- Hybrid Intelligence: Automatically switches to manual mode if CLI limits are reached.
├── automation_macOS/ # macOS Environment
│ ├── Setup_Manual.ipynb # ⚡ One-click setup
│ ├── scripts/plan_loop.py # Core AI automation engine
│ ├── prompts/ # Judge & rewrite AI personalities
│ ├── schemas/ # Quality validation rules (JSON)
│ └── INSTALL.sh # traditional installer
├── automation_windows/ # Windows Environment
│ └── [same structure]
├── .gitignore # Root level protection
└── README.md # You are here
- Fork this repository
- Create your feature branch (
git checkout -b feature/AmazingFeature) - Test ruthlessly (like our AI)
- Submit your battle-tested PR
MIT License - see LICENSE for details.
For developers who don't settle for "good enough." For teams that demand perfection. For anyone who wants their plans to actually work in the real world.
Ready to eliminate plan failures forever? Let's get started.