Codex Setup: Sol Directs Luna Max to Cut Quota, Double Output

Codex users are running a Sol-orchestrates-Luna setup: Sol creates a `luna-worker.toml` subagent in `~/.codex/agents/`, configures the model as `gpt-5.6-luna` with reasoning effort set to max, then delegates implementation to Luna while Sol handles task-splitting and code review. This division of labor is reportedly cutting quota usage while doubling output.