{"author":"jjuliano","children":[{"author":"no-name-here","children":[],"created_at":"2026-07-24T00:52:08.000Z","created_at_i":1784854328,"id":49030038,"options":[],"parent_id":49029919,"points":null,"story_id":49029919,"text":"The repo would be more helpful if it had, or linked to, any kind of quantified data on how this affects the LLM output quality or task completion.

> Why

> System prompts are 50-200k tokens. \u2026

1. Are system prompts really the primary purpose for this?

2. As of five months ago, Claude\u2019s prompt was 12-65k tokens depending on enabled tools [1]. Does \u201c50-\u201c refer to 50K? The repo should link their source for the 50-200K range.

[1] https://www.reddit.com/r/ClaudeAI/s/9OiIiHcIn3","title":null,"type":"comment","url":null},{"author":"chonghaoju","children":[],"created_at":"2026-07-24T04:14:37.000Z","created_at_i":1784866477,"id":49031169,"options":[],"parent_id":49029919,"points":null,"story_id":49029919,"text":"curious how it handles tool call schemas, those can be surprisingly token-heavy, sometimes dwarfing the actual prompt content.","title":null,"type":"comment","url":null}],"created_at":"2026-07-24T00:32:12.000Z","created_at_i":1784853132,"id":49029919,"options":[],"parent_id":null,"points":4,"story_id":49029919,"text":null,"title":"Show HN: Turo \u2013 An Aggressive Token-Saving Proxy for CLI AI Agents","type":"story","url":"https://github.com/kdeps/turo"}