FREN
Agents

Codex

Configure Codex CLI in Palabre, optionally choose a model, and recognize common errors.

Codex CLI lets you use Codex from the terminal.

Palabre can launch it as a debate agent, often as agent A to propose an initial solution.

Install before Palabre

Install and configure Codex CLI outside of Palabre. Consult the official documentation, then verify the command:

Official documentation: https://help.openai.com/en/articles/11096431-openai-codex-cli-getting-started

GitHub repository: https://github.com/openai/codex

codex --version

If Codex was installed after palabre init, synchronize the configuration:

palabre config --sync-agents

Account, model, and limits

Codex keeps its own configuration: provider, default model, sandbox, approvals, and usage limits.

Palabre only passes the prompt. If you specify --model-a or --model-b, the value is passed to the CLI via its configured model argument.

Typical configuration

"codex": {
  "type": "cli",
  "command": "codex",
  "args": ["exec", "--skip-git-repo-check", "--color", "never", "--sandbox", "read-only", "-"],
  "promptMode": "stdin",
  "shell": true,
  "role": "implementer"
}

On Windows, shell: true is often required for wrappers installed via npm or PowerShell such as codex. Claude is different in many Windows installations: claude.exe is generally called directly with shell: false.

Usage

palabre codex-claude "Critique this architecture" -t 4
palabre codex-opencode "Compare these two approaches" -t 3