OpenAI Codex CLI FAQs

Question 1

What are the different sandbox modes available in codex-cli?

Accepted Answer

The skill supports three primary modes: 'read-only' for safe file inspection, 'workspace-write' for modifying project files, and 'danger-full-access' for unrestricted system operations.

Question 2

Can I use this skill to process images or design specs?

Accepted Answer

Yes, the codex-cli includes an image flag (-i or --image) that allows you to attach screenshots or UI designs directly to your initial prompts.

Question 3

How does codex-cli handle structured data output?

Accepted Answer

Using the 'exec' command, you can pass the --output-schema flag followed by a file path to receive structured JSON responses suitable for automation pipelines.

Question 4

Why shouldn't I use sandbox flags in an interactive shell?

Accepted Answer

Sandbox modes apply OS-level filesystem restrictions that can break basic shell operations like creating temporary files; the interactive shell overlay already provides sufficient supervision.

Question 5

Can I customize the reasoning effort of the AI model?

Accepted Answer

Absolutely. You can set the 'model_reasoning_effort' to low, medium, high, or xhigh via the config.toml file or runtime flags to balance speed and accuracy.

OpenAI Codex CLI

Key Features

Use Cases

OpenAI Codex CLI

Key Features

Use Cases