How well do you know Codex?
Answer 15 multiple-choice questions on context, tools, permissions, and verification.
Good outcomes start before the prompt
The assessment looks at the operating habits that turn Codex from a code generator into a reliable engineering collaborator.
Context
Give Codex the right files, constraints, and definition of done.
Recognize when missing context makes even polished output unreliable.
AGENTS.md
Encode durable repository guidance close to the code it governs.
Know how scoped instructions shape work across directories and tasks.
Tools and permissions
Match tool access and approval boundaries to the task's actual risk.
Distinguish read-only discovery from actions that change files or external systems.
Tests and human review
Treat generated changes as a proposal that still needs evidence and judgment.
Select focused tests, inspect diffs, and keep people accountable for the final decision.
Try one question
Choose the strongest workflow, then review the reasoning behind it.
A practical path from prompting to engineering
Your level describes the habits you can apply consistently. It is not a claim about speed or seniority.
Starter
You can frame a bounded task and inspect a straightforward result.
- States the desired outcome
- Shares the relevant file or error
- Reads the generated diff
Practiced
You can guide Codex with repository context and choose useful checks.
- Adds constraints and acceptance criteria
- Uses AGENTS.md as durable guidance
- Runs focused validation
Advanced
You can delegate multi-step work while controlling risk and feedback loops.
- Scopes tools and permissions
- Breaks work into reviewable changes
- Responds to test evidence
Expert
You can design repeatable Codex workflows that improve a team's engineering system.
- Builds clear repository conventions
- Matches verification depth to risk
- Keeps final accountability with people
A result you can use on the next task
The assessment turns fuzzy confidence into specific learning priorities.
Find your baseline
See which operating habits are already reliable and which ones are still guesswork.
Name the gaps
Separate prompting issues from context, permission, and verification issues.
Improve deliberately
Use the level description to choose a focused next practice instead of collecting random tips.
Questions before you begin
What the assessment covers, how it works, and what the result means.
What does the test cover?
It covers context design, AGENTS.md instructions, tool and permission choices, testing, and human review. It focuses on how you work with Codex, not language syntax trivia.
What is the question format?
The test contains 15 single-choice questions. Each asks you to identify the strongest action or workflow for a realistic Codex scenario.
What can I do with the result?
Use it as a learning baseline. Your level highlights capability patterns and useful areas to practice, not a professional certification or hiring score.
How is the content kept current?
Questions are reviewed against publicly available OpenAI guidance and Codex CLI documentation, then revised as documented workflows and controls change.
Is Codex Level an official OpenAI product?
No. Codex Level is an independent community project and is not affiliated with, endorsed by, or operated by OpenAI.
Turn intuition into a working baseline
Fifteen questions. One clear picture of how you currently work with Codex.
Start test