在 Codex 中使用子代理和自定义代理

OpenAI Codex has officially launched its subagents feature today, following several weeks of preview behind a feature flag. Subagents are very similar to the Claude Code implementation, with default subagents named 'explorer', 'worker', and 'default'. The exact difference between 'worker' and 'default' is unclear, but based on the CSV example, 'worker' appears to be intended for running a large number of small tasks in parallel. Additionally, Codex allows users to define custom agents as TOML files in ~/.codex/agents/. These custom agents can have specific instructions and be assigned to use particular models, including gpt-5.3-codex-spark, providing flexibility for various task requirements.