Codex Path

Codex pathway 5 domains 2 × 50-item mocks Academy-aligned

This track is OpenAI certification preparation for engineers who use Codex, built from publicly available OpenAI learning objectives. It is not an official OpenAI course and not an official practice exam. Academy badges and the Codex pathway certificate of completion are not certifications and do not guarantee eligibility for a future certification.

What this track prepares you for

The Codex pathway on OpenAI Academy is three courses, all in the Build with AI category, aimed at software engineers and engineering leads:

• Get Started with Codex (~80 min) — repository tasks, making code changes, and reviewing them. The core skill is scoping a coding task, verifying the change, and giving a reviewer clear evidence.
• Extend Codex Workflows (~90 min) — turning effective practices into repeatable team workflows with clear oversight, using shared configuration, AGENTS.md, subagents, skills and plugins.
• Scale Codex Across Governed Teams and Systems (~100 min) — coordinating parallel workstreams and integrating their outputs safely, standardising configuration across many repositories, and governing the rollout.

Passing all three assessments at ≥ 80% earns the Codex pathway certificate of completion. Read the credential landscape page for what that certificate does and does not prove, and the assessment model page before your first mock.

Blueprint

Our independent mock exams weight the five domains as below. The weights are our design choice, aligned to the three courses’ published time and emphasis, not published OpenAI parameters.

#	Domain	Weight	Items (approx. of 50)	Course page
1	Codex Fundamentals and Surfaces	20%	~10	Domain 1
2	Core Coding Workflows	24%	~12	Domain 2
3	Extending and Configuring Codex	20%	~10	Domain 3
4	Team Adoption and Governance	20%	~10	Domain 4
5	Scaling Across Teams and Systems	16%	~8	Domain 5

The mindset this track rewards

Codex is an agent that writes and runs code on your behalf, so the exam repeatedly rewards one posture: you own the change, Codex drafts it, and the reviewer needs evidence. Correct answers tend to:

• Scope the task tightly and give Codex the repository context it needs — an AGENTS.md, the relevant files, the build and test commands — rather than a vague one-liner.
• Choose the surface that fits the task: CLI for scripted and non-interactive runs, IDE extension for tight edit loops, Codex cloud for long-running or parallel work.
• Use the lowest reasoning effort that gets the result, escalating Low → Medium → High → Extra High → Max → Ultra only when the task warrants it.
• Verify with tests and diffs, and hand the reviewer evidence — a passing test run, a clean diff, an auto-review summary — not just “it works”.
• Keep the agent inside a sandbox and an explicit permission mode; approve escalations deliberately.
• Standardise AGENTS.md and config.toml across repositories so behaviour is predictable at scale.

Wrong answers tend to: trust an unreviewed diff, run an agent with full network and write access by default, pick Ultra for a one-line fix, skip tests because “the model is good”, or roll Codex out org-wide without groups, roles and audit logging.

Suggested time allocation

An 18–24 hour plan for the whole track, weighted by domain and by how hands-on each one is.

Domain	Weight	Hours
Core Coding Workflows	24%	6
Codex Fundamentals and Surfaces	20%	4
Extending and Configuring Codex	20%	4.5
Team Adoption and Governance	20%	4
Scaling Across Teams and Systems	16%	3

Bias your revision toward weight × your error rate, not weight alone. Most engineers under-invest in governance (D4) and over-invest in the surfaces they already use.

Hands-on preparation checklist

Do these in a real Codex environment (CLI, IDE extension or cloud). Reading about a workflow teaches you nothing about the workflow.

• Install the Codex CLI, sign in, and run one interactive session; switch models with /model and observe the reasoning-effort levels.
• Write an AGENTS.md for a real repository: build command, test command, coding conventions, and what Codex must never touch.
• Run a scoped change with codex exec -m gpt-5.6-terra 'fix the failing test in payments' non-interactively and read the produced diff and evidence.
• Set model = 'gpt-5.6' in a shared config.toml and confirm the CLI, IDE extension and desktop app all pick it up.
• Do a review-first loop: ask Codex to plan, approve the plan, let it implement, then verify the diff and run the tests yourself before merging.
• Run the same task at Medium and at High reasoning effort and compare the result, time and cost.
• Use a git worktree so a long-running Codex task runs on its own branch without blocking your working tree.
• Turn on auto-review and read the summary Codex produces for a diff; decide whether it is enough evidence for your reviewer.
• Configure a restrictive permission mode / profile and a sandbox, then observe which actions require an approval.
• Run Codex Security (or a deep scan) on a branch and triage one finding to a fix.

Track pages