Reddit 用户称修改卡住的Claude子智能体可降低用量
一名 Reddit 用户称,他们修改了 Claude 助手智能体,让卡住的智能体直接终止,而不是无限重试。该用户表示,调整后每日用量从 30% 以上降至 10% 以下,同时称工作仍以更少 token 完成;这些说法尚未得到独立验证。
这则帖子提供了减少 Claude 工作流 token 消耗的实用经验,但相关比例和技术机制仍属于未经独立验证的个人观察。
See I’m not exactly the brightest but I found some great advice for the less neurologically inclined. Yesterday looked into the /usage for my claude, and for every single helper / subagent it was putting my entire conversation history into it (and mind you, it has 5min cache) so I was putting in 700k+ tokens into double digit agents just for simple changes.
That burns usage!
Today, I made the agents so if one’s stuck it just terminates instead of endlessly trying, burning millions of tokens. My usage dropped from 30%+ over the last few days to less than 10% today. I should be able to stretch out my remaining usage a couple days.
This is literally getting the same stuff done, just lowering token usage drastically. For the non technical doing technical work, this advice may help very much.
For you technical ones, before commenting things remember this is for the NON technical like me. Aka the dull tools in the shed.