Codex

OpenAI Codex is an agentic coding tool that runs in your terminal.

Installation

See the official OpenAI Codex documentation.

Base Invocation

codex exec "<prompt>"

Default Config

Codex is the only built-in provider that ships with a default config. When you use provider: codex, Dagu automatically applies:

skip_git_repo_check: true

This becomes --skip-git-repo-check on the CLI. You can override it by explicitly setting skip_git_repo_check: false in your step with.

Common Flags

Flag	Type	Description
--model	string	Model to use (e.g. gpt-5.5)
--full-auto	boolean	Full auto-approval mode
--skip-git-repo-check	boolean	Skip git repository validation

Example

steps:
  - name: generate-tests
    action: harness.run
    with:
      provider: codex
      prompt: "Write unit tests for the auth module"
      model: gpt-5.5
      full-auto: true

Generated invocation:

codex exec "Write unit tests for the auth module" --full-auto --model gpt-5.5 --skip-git-repo-check

YOLO Mode

Codex supports a --yolo flag that bypasses all approval prompts and sandboxing. This is useful for fully automated workflows:

steps:
  - name: auto-fix
    action: harness.run
    with:
      provider: codex
      prompt: "Fix all lint errors"
      yolo: true
      sandbox: workspace-write

Generated invocation:

codex exec "Fix all lint errors" --yolo --sandbox workspace-write --skip-git-repo-check

Warning: This disables all safety checks. Only use in isolated or trusted environments.

Fallback Example

harness:
  provider: codex
  full-auto: true
  fallback:
    - provider: claude
      model: sonnet

steps:
  - name: refactor
    action: harness.run
    with:
      prompt: "Refactor the database layer to use interfaces"

See Also

• Codex CLI Reference — Complete command and flag reference
• Non-interactive Mode — Running Codex in CI/scripts
• Config Basics — config.toml and precedence
• Sandboxing — How the sandbox works
• Models — Available model aliases and overrides