How to set the VSCode Plugin into full auto mode

Coding with ChatGPT
1

See title – that would help. Otherwise the Bot is always negging for easy tasks :wink:

And the char limit with 100 letters :melting_face:

2

Is this what you seek?

image
image279×313 7.27 KB

1 Like
3

Not really. There is something called codex --full-auto

I think.

4

Thanks for the reply, seeing codex --full-auto was of value to know what you were thinking.

Short answer:

  • Codex installed locally in the project via:
npm install --save-dev @openai/codex
  • Invoked with:
npx codex --full-auto

or

./node_modules/.bin/codex --full-auto

Disclosure:

  • I do not use Codex with TUI (text UI) in the terminal, so relied on ChatGPT 5.1 to create instructions for me, see below.
  • Only tested npx codex --full-auto did not test the option ./node_modules/.bin/codex --full-auto

Notes

  • I do use VSCode with tools such as Codex daily, do not use Text UI as I can do what I need without the level of control the TUI offers. I do use / commands as needed from the GUI.
  • While I would have expected full-auto not to request any approvals, the example slot machine project noted in the prompt below did require me to give 4 manual approvals.
Prompt for ChatGPT 5.1

The prompt that follows is very specific to my workflow, adjust as needed for your workflow.

Please provide detailed, step-by-step instructions—suitable for an experienced developer on Windows 11—demonstrating how to use codex --full-auto inside a Projects folder (with full read/write access).

Requirements

Terminals

  • Use Git Bash for all commands unless a specific step requires another terminal.
  • If any command must be run in PowerShell or CMD, explicitly state so; otherwise assume Git Bash is used for everything.

Codex Installation & Configuration

Provide clear explanations for:

  • How to install Codex on Windows 11
  • How to install or reference its dependencies
  • How to run Codex without adding it to the Windows PATH (e.g., using absolute paths or local project-level installation)
  • How to configure Codex so that codex --full-auto works inside a project folder
  • How to verify Codex is operational (codex --help, version checks, etc.)

Command Verifications

Do not recommend modifying system PATH.
Instead:

  • Show how to verify whether needed tools exist (git, grep, etc.)
  • Show how to run required executables using explicit paths if necessary
  • Include test commands confirming that each required tool runs successfully in Git Bash

Project Walkthrough

After environment setup is confirmed, walk through:

  1. Creating a new project (e.g., ~/Projects/slot-machine-demo).

  2. Running codex --full-auto to generate a simple web-based slot machine game (HTML/CSS/JS).

  3. Any interaction needed with Codex during the generation process.

  4. Using Git to initialize the repo, stage files, and commit changes — no GitHub pushes.

    • No need to explain basic Git commands; assume developer-level knowledge.

  5. Any verification steps (viewing files, confirming output, confirming commits).

Experienced Developer Note

Skip explaining what common commands (like git add, cp, ls) do.
Only environment-specific instructions (Windows quirks, Codex install, absolute paths, etc.) should receive detailed explanation.

Before beginning, ask any necessary clarifying questions (e.g.,

  • installation method preferred for Codex,
  • whether the user wants a local project-specific installer,
  • exact location of Codex executables if already downloaded,
  • desired HTML/JS structure for the slot machine game).
5

Well, there you see the issue, it needs to be auto. That is something the OpenAI team should add – an auto mode for grunt work.

1 Like
6

After reading your last reply what you asked seems so obvious that surely it must exist.

Since OpenAI does have a GitHub Codex repo did a search

repo:openai/codex full-auto

Code search results · GitHub

At this point I don’t plan to test out options based on what is presented in the search as I work on Windows and don’t use WSL as it will allocate GBs of memory that are then unavailable for the Windows OS.

Thanks for asking the question.

7

After looking at the agent’s working process, I found a solution. You need to write an ExecPlan that is loaded over the PLANS.md file. That actually works quite well. However, it would be easier if there was an option to change the agent’s default plan/action to recognise certain prompts. But I assume this is actually wanted, or let’s call it a feature. :wink:

1 Like
8

But Codex-CLI is not a VSCode plugin? :thinking: