Use Codex Subagents

Orchestrate Codex subagents for the requested task. Keep the skill and custom-agent identity stable across worker-model upgrades. The bundled TOML defines the worker defaults; an explicit worker model or reasoning effort in the user's request overrides the matching default. Parent model and reasoning settings remain outside this managed profile.

Execution contract

• Use at least one subagent unless multi-agent tooling is unavailable or doing so would violate a safety or permission boundary.
• Do not skip delegation merely because the parent could complete the task alone. For a small or tightly coupled task, use one subagent as the focused executor or independent verifier.
• The parent remains responsible for decomposition, user-facing decisions, synthesis, verification, and the final answer.

Preserve the parent runtime

• Leave the parent model and reasoning effort exactly as selected for the current Codex CLI or ChatGPT Desktop Work mode session.
• Never set, override, or persist a parent model or model_reasoning_effort on behalf of this skill.
• Do not change model or reasoning settings in $CODEX_HOME/config.toml, a profile, command-line options, application settings, or the active parent session.
• Do not derive the subagent model or effort from the parent, and do not apply the subagent settings to the parent.
• Apply the resolved worker model and reasoning effort only to workers spawned by this skill.

Resolve the worker profile