Codex Core Concepts | Beginners Tutorial

Before using Codex, it's important to understand several core concepts. These concepts form the foundation of how Codex works.

Prompt

You interact with Codex by sending Prompts, describing the tasks you want it to complete.

Prompt Working Loop

After you submit a Prompt, Codex works in the following loop:

Call the language model to understand the task
Execute operations indicated by the model output (read files, edit code, run commands)
Feed operation results back to the model
Loop execution until the task is complete or you cancel

Codex adopts an Agent loop working mode—the model outputs operation instructions, operation results are fed back to the model, and this cycle repeats until the task is complete.

Principles of Effective Prompts

Principle	Description	Example
Include verification steps	Codex outputs higher quality when it can verify its work	"Write a function with test cases to verify it handles empty lists"
Break down complex tasks	Small tasks are easier to test and review	"Step 1: Create the model; tell me when done before continuing to Step 2"
Provide context	Reference relevant files and images	"Referencing the style of src/auth.py, implement similar functionality"

Codex performs best when processing small, focused tasks. Large tasks should be broken down into small steps and executed gradually.

Thread

Thread is a single task session: your Prompt plus subsequent model outputs and tool calls.

Thread Types

Type	Runtime Environment	Characteristics
Local thread	Your machine (within sandbox)	Can read/write files, use existing tools, execute commands
Cloud thread	Cloud isolated environment	Clone repository to run, suitable for parallel tasks, delegate across devices

Thread Usage Rules

Running threads can be concurrent, but note:

Avoid two threads modifying the same file simultaneously
Threads can be resumed later by continuing with another Prompt
Long-running tasks may automatically compress context

Context

When you submit a Prompt, it includes context that Codex can use—references to relevant files and images.

Context Sources

IDE extension: Automatically includes list of open files and selected text ranges
Manual specification: Reference file paths or attach images in the Prompt
Conversation history: Previous conversation content in the thread

Context Window

All information in a thread must fit within the model's context window.

Codex monitors and reports remaining space. When approaching the limit, you will be prompted.

Automatic Compact

For longer tasks, Codex may automatically compress context.

The compression mechanism summarizes relevant information, discards less important details, and frees up space to continue processing.

Managing Context

# Start new session to release context
/new

# View current context usage
/status

When Codex reports high context usage, consider starting a new session or reducing historical messages.

Sandbox

Sandbox is Codex's security isolation mechanism, preventing accidental modification of files outside the workspace.

Sandbox Modes

Mode	File Modification	Network Access	Use Case
Read-only	Prohibited	Prohibited	Read-only analysis, code review
Workspace-write	Workspace only	Prohibited	Daily development (default)
Full-access	Allowed	Allowed	Fully trusted environment (use with caution)

Approval Mechanism

Certain operations require your confirmation before execution:

Executing Shell commands (especially rm, kill, etc.)
Modifying or deleting files
Accessing sensitive directories (such as ~/.ssh/, /etc/)
Network requests

Setting Sandbox Mode

# Switch sandbox mode in CLI
 codex --sandbox workspace-write

# Or specify in Prompt
"Analyze this code, don't modify any files"

Sandbox is the first line of defense in Codex's security strategy, ensuring AI operations don't exceed expected scope.

Approval Policy

Approval policy controls whether Codex requires confirmation before executing operations.

Policy Types

Policy	Description	Behavior
`ask`	Ask every time	Request confirmation before sensitive operations (default)
`approve`	Auto-approve	Execute automatically without confirmation (use with caution)
`deny`	Auto-deny	Reject all operations that may have side effects

Configuring Approval Policy

# ~/.codex/config.toml

approval_policy = "ask"

Summary

After understanding these core concepts, you can use Codex more effectively:

Prompt: Clearly describe tasks, include verification steps
Thread: Manage task sessions, note concurrency rules
Context: Provide relevant context, monitor window usage
Sandbox: Understand security mechanisms, choose appropriate modes

FAQ

Q: Should Prompts be in Chinese or English?

Codex supports multiple languages, but English usually works best. When using Chinese, ensure descriptions are detailed and explicit.

Q: How to check current thread status?

Use the /status command to view thread ID, context usage, and configuration information.

Q: What to do when context window is full?

Start a new session (/new), or let Codex automatically compress historical records.

Q: Can sandbox mode be switched dynamically?

Can be specified when starting CLI, or set default values in configuration file.

YouTip

Codex Concepts