Token Saver

Token Saver helps users request concise, useful AI responses and reduce avoidable context in prompts, summaries, and technical material. Its focused skills cover concise response preferences, safe summarization and compression planning, and an evidence-aware token-efficiency review. It preserves negation, numbers, units, code, commands, paths, error text, uncertainty, and safety-critical detail; asks before removing information that could change a decision; and explains tradeoffs. It is an independent text-guidance plugin inspired in part by the public Caveman project (Julius Brussee), not an official Caveman product or executable port. It does not install or run Caveman, proxy model traffic, inspect agent history, count actual model tokens, guarantee savings, or change billing. Results depend on the model and task; verify important technical or safety information against its source.