codex-config_skill

Overview

This skill manages Codex CLI configuration files and related runtime settings. It helps set up model options, sandbox policies, profiles, MCP servers, and runtime overrides to control behavior and safety for agent workflows.

How this skill works

The skill reads and writes the primary config at ~/.codex/config.toml and obeys a precedence list of skills paths (project, repo, user, system, built-in). It exposes structured TOML blocks for model settings, approval_policy, sandbox_mode, profiles, MCP servers, and convenience CLI overrides to change values at runtime.

When to use it

• Setting up or editing ~/.codex/config.toml for local or project-specific settings
• Creating or switching named profiles for different workflows (fast vs thorough)
• Configuring sandbox_mode or approval_policy before running potentially risky operations
• Registering or troubleshooting MCP servers and their environment keys
• Overriding config values on the command line for temporary runs

Best practices

• Keep a minimal ~/.codex/config.toml and use project-level .codex/skills/ for per-repo rules
• Define named profiles for repeatable workflows rather than changing global defaults
• Use approval_policy and sandbox_mode conservatively in shared environments
• Validate TOML after edits (toml-lint) and test overrides with codex -c
• Enable MCP servers only when tested; store credentials in environment blocks, not inline secrets

Example use cases

• Create a max profile for deep reasoning and a fast profile for quick iterations, then run codex -p max "refactor module"
• Switch to workspace-write sandbox for trusted project work or set read-only for auditing sessions
• Add an MCP server entry to run local toolchains with command and env keys, then verify with codex mcp list
• Temporarily force a different model on the CLI via codex -c model="gpt-5.2-codex" for A/B testing
• Use convenience flags like --full-auto to combine approval and sandbox settings for hands-off runs