Codex Prompt Generator and AGENTS.md Builder

This free browser tool writes AGENTS.md files, task briefs and code review rules for Codex, OpenAI's coding agent. It recommends GPT-6.1 Sol where available and helps set effort for the task. No sign-up is needed. It generates instructions without calling a model.
Models last verified:

Generate your prompt

Access 40,000+ AI Prompts for ChatGPT, Claude & Codex — Free!

Subscribe to get instant access to our complete Notion Prompt Library — the largest curated collection of prompts for ChatGPT, Claude, OpenAI Codex, and other leading AI models. Optimized for real-world workflows across coding, research, content creation, and business.

Access Free Prompt Library

Which model and reasoning effort should I use with Codex?

The table below maps effort levels to task types, with GPT-6.1 Sol recommended where available.

TaskReasoning effortWhy
Small, well-scoped change or questionLowFast answers with enough reasoning for clear tasks
Bug fix, feature or refactor across a few filesMediumThe usual setting for everyday Codex work
Complex change, ambiguous problem or migrationHighMore thinking time before Codex edits code
Long, reasoning-heavy taskExtra highThe most thinking time; slowest and most expensive
Model: GPT-6.1 Sol where available. Effort levels from OpenAI’s Codex guidance, checked 17 October 2026.

How do I use my prompt?

  1. Choose task brief, AGENTS.md builder or code review rules.
  2. Add the relevant goal, project details, commands and boundaries. Include reproduction steps and errors for bug fixes.
  3. Generate the text and check that the instructions match your repository.
  4. Paste task briefs into Codex. Save AGENTS.md in your repository root, and paste review rules into its Code Review Rules section.

Prompting tips for Codex

  • State the goal, relevant context, output, boundaries, and the final check that defines success; use only the pieces the task needs. OpenAI: Prompting
  • For code work, name the intended behavior, point to relevant code or reproduction steps, preserve constraints, and say how to verify the change. OpenAI: Prompting
  • For a reproducible bug, give the exact reproduction recipe and constraints; after the fix, rerun that recipe. OpenAI: Prompting
  • Ask Codex to plan before editing when a task is complex, ambiguous, or difficult to describe. Codex best practices
  • Use low effort for well-scoped work, medium or high for harder changes or debugging, and extra high for long agentic reasoning. Codex best practices
  • Keep AGENTS.md short and accurate, covering practical commands, conventions, constraints, and verification rather than vague rules. Codex best practices

Worked examples

Bug fix brief

Before: Fix the checkout total ignoring discount codes when the cart has more than one item.

After (made with this generator):

Goal: Fix the checkout total so it includes the discount code when the cart has more than one item.

Context: The bug is in @src/cart/totals.ts. To reproduce: add two items, apply code [CODE], open checkout. The total ignores the discount.

Boundaries: Do not change the public interface of @src/cart/index.ts. Add no new dependencies.

Output: A focused change with a test that fails before the fix and passes after it.

Done when: the reproduction steps show the discounted total, the smallest relevant test suite passes and lint is clean. Re-run the reproduction steps after the fix.

GPT-6.1 Sol, effort Medium. Start with /plan.

What the generator added: The prompt adds the relevant file, reproduction steps, interface and dependency boundaries, and a regression test requirement. It also requires lint, the smallest relevant test suite and a repeat of the reproduction steps.

Starter AGENTS.md

Before: Write an AGENTS.md file for this repository.

After (made with this generator):

# AGENTS.md

## Project
[One line: what this repository is, language and package manager.]

## Commands
- Install: [INSTALL COMMAND]
- Fast tests: [COMMAND FOR THE SMALLEST RELEVANT TESTS]
- All tests: [FULL TEST COMMAND]
- Lint: [LINT COMMAND]

## Conventions
- Money values are whole numbers of pence, never decimals.

## Boundaries
- Never edit generated files in [FOLDER].
- Ask before adding a dependency.

## Code Review Rules
- Flag money calculations that use decimals; the safe path is whole pence.
- Flag new API routes without an authentication check, unless listed in [PUBLIC ROUTES FILE].


Save as AGENTS.md in the repository root.

What the generator added: The file adds sections for project details, commands, whole-pence conventions and boundaries on generated files and dependencies. Review rules flag decimal money calculations and new API routes without an authentication check, with an exception for listed public routes.

Frequently asked questions

What is AGENTS.md?

AGENTS.md is a file of standing instructions that Codex reads in a code repository. It can describe repository layout, commands, conventions, constraints, verification criteria and code review rules.

Where should AGENTS.md go?

Put AGENTS.md in the repository root for project-wide instructions. Project files apply from the root down, and a file nearer the working folder wins. A personal file in the .codex folder applies to every project.

How long should AGENTS.md be?

Keep it short and accurate, well below Codex’s default 32 KiB limit, controlled by project_doc_max_bytes. Include commands, conventions and constraints that matter to the repository. Avoid long lists of vague rules.

Which model and effort should I use?

Use GPT-6.1 Sol where available. Choose Low effort for well-scoped tasks, Medium or High for complex work, and Extra High for long, reasoning-heavy tasks. Use /plan before editing for multi-step, complex, ambiguous or difficult-to-describe tasks.

How do I start a Codex review?

Use /review inside Codex. On GitHub, add a pull request comment containing @codex review or @codex security review; GitHub reviews flag only P0 and P1 issues. Review rules do not replace tests, branch protections or required approvals.