Codex Tutorial Path: Execution Surfaces and Review Boundaries
The same Codex task has different file access, tools, feedback timing, and review needs on different execution surfaces. Choose the place where work runs before you optimize the work itself.
Start with the problem you have
Section titled “Start with the problem you have”| What is happening now | Read first | Your first verifiable result |
|---|---|---|
| You do not know whether to use the CLI, IDE, desktop, or Cloud | The four Codex execution surfaces | A task-to-surface matching table |
| You want one safe small change from the terminal | The Codex CLI workflow | A local delivery with a diff and checks |
| You are unsure about sandbox, network, and permission boundaries | Sandbox and permissions | A minimal authorization list for the task |
A reusable learning route
Section titled “A reusable learning route”- Choose a surface by local feedback, async waiting, code access, and risk level instead of choosing the tool that seems strongest.
- State which commands and external actions each run may take, and which ones must stop for confirmation.
- Use the boundaries of automated review to separate finding a problem from merging, publishing, or deleting data for someone else.
What to keep after reading
Section titled “What to keep after reading”- A record of the task, execution surface, permissions, and verification method.
- One reviewable diff, command output, and unresolved risk.
- A rule that can stop clearly when human review or external authorization is needed.
Continue from here
Section titled “Continue from here”If you want to put Codex into a fuller AI coding loop, return to the AI Coding Practice Path. When long tasks belong in the Cloud, read which tasks fit Codex Cloud.