9 Claude Prompt Templates That Reduce an 8‑Hour Workday to 47 Minutes of Manual Work
The author built nine Cowork slash‑command prompt templates for Claude, measured a median of 47 minutes of manual keyboard time versus an 8‑hour day, and demonstrated a real‑world weekly saving of 34 hours by automating daily, weekly and research tasks with explicit termination rules and role definitions.
Overview
In a full 8‑hour workday the author’s median manual keyboard time was 47 minutes. By converting all recurring tasks to background automation using nine Cowork slash‑command prompt templates, the total weekly manual effort fell by 34 hours, based on a comparison of April (no templates) and May (with templates) logs.
1. Daily Brief (47 min → 4 min)
Runs at 07:30, aggregates overnight emails, messages, calendar, holdings and news, then classifies items into "Urgent / Can‑Delay / Irrelevant". The prompt is defined as:
/morning-brief
角色:我的首席助理。读取我夜间接收的所有信息,整理成一页简报。
拉取数据源:
- 邮箱:过去12小时,收件箱+已发送(确认我曾承诺的事项)
- 企业微信/钉钉私信:过去12小时,@提及+私信,指定群聊=[列表]
- 日历:今日+明日上午,包含参会人背景信息
- 持仓平台:当前持仓、盈亏与入场价对比
- 新闻:[3个RSS源/资讯网站],过去12小时,仅筛选含[关键词]的条目
输出格式(Markdown,最多一页):
## 10点前需回复
[项目符号:发件人 + 1行概述 + 建议动作]
## 可延后处理
[项目符号,最多5条]
## 无效信息(仅标记已查看)
[每条1行,不展开细节]
终止条件:处理完全部5个数据源,且内容不超过400字后停止。无需对简报再做概括。Key effect: The termination condition limits output to one page, cutting the original 1400‑word draft by 60 % without losing information.
2. Competitor Scan (3 h → 18 min)
On‑demand batch extraction of competitor websites, pricing, blogs, social posts, hiring and financing, then cross‑compares with the author’s positioning.
/competitor-scan
输入项:
- 竞品列表:[逗号分隔,最多8个]
- 自身定位文档:[云盘路径]
- 聚焦维度:[定价 | 产品 | 招聘 | 市场策略 | 全维度]
执行流程:
1. 对每个竞品抓取:官网首页、定价页、最近5篇博客/公众号文章、主账号近30天社交动态、公开招聘岗位。
2. 将竞品定位与自身定位文档交叉对比。
3. 标注近60天内与我方定位趋同的竞品。
4. 标注近90天内的招聘/融资信号。
输出格式(Markdown):
## 核心结论(最多3行)
## 各竞品详情
- 定价与我方的差异
- 定位差异(近60天)
- 招聘信号
- 融资信号
- 潜在动作预判(仅基于现有信息,不推测)
终止条件:填完每个竞品的5个字段后停止。不写行业概览,不预测未来。Key constraint: Prohibits future predictions; each run saves roughly 40 minutes.
3. Email Triage & Draft (90 min → 11 min)
Runs at 09:00, 13:00 and 17:00, classifies incoming mail, generates reply drafts in the author’s tone, and tags them for review.
/triage
输入项:
- 时间范围:[上次运行 | 早间 | 24小时内]
执行流程:
1. 拉取指定时间范围内的所有邮件,排除广告/系统通知。
2. 邮件分类:需回复 | 仅参考 | 已处理。
3. 为“需回复”邮件生成草稿:
- 参考我发给同一人的最近5封已发送邮件(匹配语气)
- 无历史往来时,默认模板:确认收到 + 1个核心问题(共3句话)
4. 在邮箱中保存草稿,标注“AI草稿”标签,不自动发送。
5. “仅参考”类邮件按发件人聚合,每人1行概述。
输出格式:
## 需我确认(草稿已生成)
[列表:发件人 + 主题 + 草稿核心内容]
## 仅参考
[按发件人分组]
## 已处理/闭环
[每条1行]
终止条件:所有未读邮件分类完成后停止。不为已生成草稿的邮件重复创作。Core detail: Uses the last five sent mails to match tone; forbids over‑generation.
4. Meeting Prep (30 min → 3 min)
Two hours before an external meeting, pulls participant history, documents, social updates, and lists unfinished items, then generates three core questions.
/meeting-prep
输入项:
- 会议:[日历事件ID | 下一场外部会议]
执行流程:
1. 识别外部参会人(非本组织人员)。
2. 为每位参会人拉取:最近10封邮件(双向)、共享文档、最近1次领英更新、近30天社交动态。
3. 梳理上一次沟通内容及未完成事项。
4. 基于未完成事项生成3个核心开放问题。
输出格式(单页):
## 参会人员
[每位参会人:2行背景信息]
## 上次沟通要点
[时间、达成共识、未完成事项]
## 3个核心问题
## 近期关键变化
[仅保留影响本次会议的内容]
终止条件:内容填满1页后停止,超出部分直接删除。Design logic: One‑page concise output, avoids four‑page redundancy.
5. Weekly Status Report (2 h → 7 min)
Every Friday at 16:00 aggregates tasks, docs, chats, calendar data and produces a tailored report for internal teams, clients or board.
/weekly-status
输入项:
- 周期:[本周 | 上周]
- 受众:[内部团队 | 客户 | 董事会]
执行流程:
1. 拉取本周已关闭的任务工单(含PR标题、审核人)。
2. 拉取本周创建/编辑的笔记文档。
3. 读取本周工作群(进度/交付/设计)的聊天摘要。
4. 日历统计:外部会议、内部会议、深度工作时长。
输出格式(Markdown,按受众调整长度):
- 内部团队:1页
- 客户:1.5页,弱化失败表述,突出风险点
- 董事会:2页,含“指标vs目标”表格
固定板块:
## 已交付事项
## 进行中事项
## 阻塞/风险点
## 核心指标(仅董事会版)
## 下周重点
终止条件:达到对应受众的长度后停止。不编造无数据源支撑的指标。Most important rule: Never fabricate data; mark missing data as “未统计”.
6. Document Review with Q&A (90 min → 9 min)
Handles PDFs or long texts, extracts arguments, identifies gaps, and generates five structured questions with answers.
/doc-review
输入项:
- 文档:[文件路径/链接]
- 参考上下文:[我的相关笔记路径,可选]
- 审阅深度:[快速浏览 | 常规审阅 | 深度分析]
执行流程:
1. 通读全文,识别:核心论点、关键主张、支撑证据、证据缺口。
2. 若提供参考上下文,标注与我过往写作的冲突/一致点。
3. 生成5个核心问题(模拟认真读者视角),用文档内容作答(无相关内容则写“未提及”)。
输出格式:
## 核心论点(1行)
## 关键主张(3-5条,标注证据可信度)
## 不一致/缺口点
## 与我过往内容的冲突/契合(如有)
## 核心问答(5个)
## 结论:全文阅读 | 快速浏览 | 无需阅读
终止条件:按上述结构完成后停止。不按章节逐段复述文档。Core value: Replaces verbose chapter‑by‑chapter summaries with a 30‑second focused Q&A.
7. Portfolio Audit (45 min → 3 min)
Runs three times daily, links news and volume data, flags positions needing manual intervention, but never recommends trades.
/poly-audit
输入项:
- 钱包地址:[我的持仓钱包地址]
- 预警阈值:[10% | 20% | 30%]
执行流程:
1. 拉取所有持仓:入场价、当前价、盈亏额。
2. 为每个持仓检索过去12小时的相关新闻事件。
3. 与[我的关注清单]交叉核对。
4. 标注满足以下条件的持仓:
- 盈亏幅度超过阈值
- 新闻提及底层事件
- 该市场近6小时成交量增幅超50%
输出格式:
## 需重点关注
[持仓名称 + 原因 + 建议动作]
## 持有观望
[每条1行]
## 自上次审计后已平仓
[每条1行,含实际盈亏]
终止条件:检查完所有持仓后停止。不推荐新仓位,不预测市场方向,仅做审计。Hard constraint: Pure audit; AI must not suggest new positions.
8. Deep Research (4 h → 28 min)
Spawns five parallel sub‑agents to collect academic, news, social, primary documents and opposing viewpoints, then synthesizes a concise brief.
/research-deep
输入项:
- 研究主题:[自由文本]
- 深度:[概览 | 简报 | 完整分析]
- 时间限制:[30分钟 | 60分钟 | 120分钟]
执行流程(启用子代理并行处理):
生成5个并行子代理:
- 代理A:学术来源(知网、arXiv、核心期刊)
- 代理B:新闻+行业媒体(近12个月)
- 代理C:社交平台讨论(该主题头部账号,近90天)
- 代理D:一手文档(政府备案、企业公告、公开数据集)
- 代理E:反方观点:3个核心反对者及核心论点
每个子代理输出:5个来源 + 3行核心发现。
协调器合成最终内容:
## 核心结论(3行,面向专业质疑者)
## 主流共识
## 反方观点
## 证据不足的开放问题
## 基于证据的建议
终止条件:达到时间限制后停止。每个子代理仅运行一轮,完成后不再迭代。Efficiency essence: Multi‑threaded parallel execution with explicit termination to avoid endless retrieval.
9. Content Repurposing (90 min → 12 min)
One‑click adaptation of a long article to Weibo, LinkedIn, public account, WeChat group and email, each with distinct opening hooks and tone.
/repurpose
输入项:
- 原文:[长文文件路径/链接]
- 目标平台:[微博 | 领英 | 公众号 | 企业微信群 | 邮件 | 全部]
- 语气:[我的默认风格 | 正式 | 口语化]
执行流程:
对每个选中平台:
1. 提取原文最核心的主张。
2. 按平台规则改写:
- 微博:6-10条系列内容,单条<280字,首条需吸睛钩子
- 领英:1条帖子,300字,专业语气
- 公众号:2段摘录 + 1行引导回原文的CTA
- 企业微信群:4行概述 + 链接,适配分享群调性
- 邮件:60字短文 + 标题 + 行动指令
约束条件:
- 不同平台不使用相同开场
- 单次引用原文不超过15字
- 参考[我该平台近10条内容]匹配语气(如有)
输出格式:按平台分区块,内容可直接复制粘贴。
终止条件:所有选中平台改写完成后停止。不生成额外“附赠版本”。Key rule: Each platform must have a unique opening to prevent content‑pipeline homogenization.
Common Characteristics of the Nine Templates
Explicit, quantifiable termination conditions that bound execution time and output size.
Fixed structured output sections that prevent free‑form drift.
Role definition at the first line (e.g., chief assistant, audit officer, research coordinator) to constrain AI scope.
Two Failure Modes Observed
Missing clear stop rules leads to endless iteration and bloated output (14 of 21 discarded templates suffered this).
Task‑boundary creep when the role is undefined, causing the AI to overstep (e.g., email templates predicting reply rates, audit templates recommending trades).
Deployment Steps
Create each template as a Cowork slash‑command; initial setup takes about 10 minutes.
Replace bracketed placeholders with personal data sources (email, cloud storage, calendar, news feeds).
Run the full set for one week without adding extra features; delete unused or empty sections.
Review logs after a week, make minor tweaks, and then the system stabilizes for long‑term reuse.
Time‑Saving Calculation
Each template’s saved time × its execution frequency yields a theoretical maximum of 72 hours per week.
After accounting for overlapping data sources, prompt fine‑tuning time, and concurrent background work, the net weekly saving is 34 hours.
First week saved ~18 hours; after optimization the stable saving is 34 hours, showing that prompt refinement is more impactful than the raw template.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
AI Architecture Hub
Focused on sharing high-quality AI content and practical implementation, helping people learn with fewer missteps and become stronger through AI.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
