用量收紧后把小 Bot 团队再瘦身：10 步审计清单 + 一段只读审计提示词
Making a small bot team leaner after the usage squeeze: a 10-step audit plus a read-only audit prompt
X, @jakeinthesz, 2026-11-10
编程与仓库 全文核对建议先读
https://x.com/jakeinthesz/status/2120163569392341486
场景
作者主要把订阅用在 Cursor 写代码，Grok Bot 只是副业项目的小 Bot 团队；用量调整后第一次提前撞线
Most of the author's plan goes on coding in Cursor; Grok Bot runs a small bot team for a side project, and after the usage change he hit his limit early for the first time
他们怎么做
① 审计每个 Bot、Routine、笔记：找过期事实、重复劳动、过度建设、没人读的输出；② 定标准：公开内容先草稿再人批，每个事实带来源文件或日期，失败要大声、平时安静；③ 精简描述（每次聊天和运行都会读），只留身份、规则和指针；④ 重复方法写成共享 Skill，描述里只放一行指针；⑤ 缩小开场读取：共享文件只留现状，其余归档或用脚本只打印所需；⑥ 固定检查用脚本、定时判断用 Routine、无连接器/API 才用浏览器；⑦ 角色清晰、一名一职，一个主 Bot 派活并发日报；⑧ 笔记自维护：每次运行归档已解决项、带来源与日期修正过期事实，每月在现有 Routine 内做 90 天清理；⑨ 修剪 Routine：合并重叠，每小时→每天、每天→每周两次，重扫描放每周，没人看的 Bot 暂停，内容最多排 7 天；⑩ 批量对话后再测量：每个 Bot 一次合并消息，按 Bot/Routine/天看用量，不宣称数据没显示的节省
1) Audit every bot, routine and note for stale facts, duplicated work, overbuilding and unread output; 2) set standards: public work is drafted then human-approved, every fact has a source file or date, fail loudly otherwise stay quiet; 3) slim descriptions (read on every chat and run) to identity, rules and pointers; 4) write shared skills for repeated methods with a one-line pointer in each description; 5) shrink what's read at start: current facts only, archive the rest or let a script print what's needed; 6) scripts for fixed checks, routines for scheduled judgement, browser only without a connector or API; 7) clear roles, one name each, one lead bot dispatches and sends a daily briefing; 8) self-maintaining notes with a monthly 90-day cleanup inside an existing routine; 9) trim routines: merge overlaps, hourly to daily, daily to twice weekly, heavy scans weekly, pause unread bots, plan content at most 7 days ahead; 10) batch chats, then measure usage by bot, routine and day, never claiming savings the data doesn't show
可以拿走的做法
原文末尾的提示词可直接用：让 Bot 按 10 步审计、按影响排序提方案、等确认再改、只用真实用量数据
The closing prompt is reusable as-is: have the bot audit against the 10 steps, propose changes biggest first, wait for OK before changing anything, and use real usage data only结果与限度
作者称用量变化后「没有崩」，但未给具体节省数字；配图内容未读到；6 条回复未逐条核对。
The author says the setup didn't break after the usage change but gives no savings figures; the attached images were not read; the 6 replies were not reviewed one by one.
来源
- 平台
- X
- 作者
- @jakeinthesz
- 日期
- 2026-11-10 22:36 Asia/Taipei (UTC+8)
- 日期依据
- 案例库画布，采集截止 2026-11-11 09:26 Asia/Taipei
- 时间窗
- 近7天 案例字段「近7天」相对采集截止 2026-11-11 09:26 Asia/Taipei，不是相对打开页面的今天
- 原分类
- 一手实践/用量优化
这条证据标签是什么意思
全文核对
来源已经读过（帖子、文章或字幕）。这只说明原文被读过，不代表本站复现了作者的结果。
标签不是评分。全文核对只表示来源被读过，不表示结果已被复现。