tokensavings-pro splits Sol, Luna work

The clever part is model specialization, not a quota exploit: it may stretch usage for delegation-friendly coding tasks, but it cannot alter plan-level limits.

• –Sol stays out of the repository while Luna handles implementation, testing, and verification
• –The handoff packet reduces expensive reasoning-model context, but the project provides no benchmark proving 5x total-token savings
• –Official plan limits and model allowances still apply; the repo’s 20x figure refers to the Pro 20x tier’s Luna allowance versus Plus
• –A five-handoff cap and disagreement stop condition add useful guardrails against runaway orchestration
• –Independent testing of similar token-saving modes found inconsistent savings and some regressions in robustness, so developers should validate quality on their own workloads