OMX Prompts

Autonomous product-delivery prompts for OpenAI Codex + oh-my-codex.

Build from empty folders, ship from docs, improve existing apps, redesign frontends, harden QA, review security, and produce final evidence.

Quick Start · Choose A Prompt · Examples · Full Guide

OMX Prompts is a curated set of long-form operating prompts for running Codex like a small autonomous product team.

The prompts are designed to make Codex:

• inspect the local project before asking questions
• interview you only when the goal is genuinely under-specified
• create product/design/architecture/test documentation when needed
• use Team mode for parallel specialist lanes
• use Ralph-style loops for persistent fix/verify work
• run UltraQA before final completion
• produce final reports with verification evidence

If this saves you setup time or helps Codex ship more complete work, a star helps other builders find it.

Install Codex and OMX:

npm install -g @openai/codex oh-my-codex
omx setup --scope user --merge-agents
omx doctor

Start from a project folder:

cd /path/to/project
git init
codex

Paste the most universal prompt:

prompts/omx-super-universal-autonomous-delivery-prompt.md

Fill only the important placeholders. Leave the rest as AUTO_DETECT, AUTO_DECIDE, or UNKNOWN if you want Codex to infer from context.

Need the complete setup guide? Read INSTRUCTIONS.md.

For a more detailed decision tree, see docs/PROMPT_SELECTION_GUIDE.md.

prompts/omx-super-universal-autonomous-delivery-prompt.md is the default choice.

It handles:

• empty folder -> interview -> documentation -> implementation
• docs-first project -> full implementation
• existing app -> feature/module/integration/redesign
• maintenance work -> bugfix, QA, security, performance, production, release
• unclear request -> inspect first, then ask one focused question at a time

flowchart LR
  A["Project folder"] --> B["Detect state"]
  B --> C["Interview if needed"]
  B --> D["Read docs/code"]
  C --> E["Plan + design + architecture"]
  D --> E
  E --> F["Team lanes"]
  F --> G["Implement"]
  G --> H["Tests + build + browser/API checks"]
  H --> I["Ralph-style fix loop"]
  I --> H
  H --> J["UltraQA"]
  J --> K["Final report with evidence"]

Filled placeholder examples live in examples/:

• empty-project-to-product.md
• docs-to-full-product.md
• existing-app-feature.md
• frontend-redesign.md
• api-integration.md
• qa-hardening.md
• security-hardening.md

A good Codex/OMX run should end with:

• working code
• updated docs
• tests or justified test gaps
• build/lint/typecheck evidence
• browser/API verification where relevant
• QA findings and fixes
• security review where relevant
• FINAL_REPORT.md or task-specific report with traceability

If the result is only a plan, scaffold, fake UI, dead links, or vague “done” summary, the prompt did not finish the job.

• Request a new prompt with the Prompt request issue template.
• Improve an existing prompt with the Prompt improvement template.
• Share real results with the Showcase / field report template.
• Join the welcome discussion: Welcome to OMX Prompts.

These prompts give Codex a lot of autonomy. Use them inside git repositories and commit important work before long autonomous runs.

The prompts are designed to ask before:

• destructive actions
• production deployment
• paid external service usage
• missing credentials
• irreversible product/business decisions

MIT. Use, modify, and adapt these prompts for your own projects.