ThePlant Workflow Runner

A tool designed to maximize AI productivity by enabling autonomous, long-running task execution without human intervention.

The Problem

AI Efficiency Challenge

The key to maximizing AI productivity is enabling it to work independently for extended periods. The longer an AI can work autonomously without requiring human input, the more efficient it becomes.

However, current AI assistants have a fundamental limitation: when given large amounts of content or instructions, they don't execute everything meticulously. Instead, they:

• Pick out what they perceive as "key points"
• Skip over details they consider less important
• Execute only a subset of the full instructions

This is the opposite of what humans want. When you delegate work to another person, you can hand them a comprehensive document and expect them to work through it from start to finish. But even humans have the same tendency - they might not follow every instruction, might cherry-pick the easy tasks, or skip what seems unimportant.

The Solution

To make AI execute your plan meticulously and completely, you need to:

1. Break down tasks into individual units
2. Feed tasks to the AI one at a time
3. Only provide the next task after the current one is completed

This approach ensures:

• Every task gets full attention
• No instructions are skipped or overlooked
• Each request to the AI is fully utilized
• The AI runs until it hits its maximum conversation limit (the "Continue" button appears in Windsurf)

How It Works

Workflow Definition

Create workflow files in .windsurf/workflows/ directory using Markdown format. Each workflow contains steps that the runner will extract and feed to the AI one by one.

Workflow Format

---
description: Brief description of what this workflow does
---

## How to Execute This Workflow

**⚠️ MANDATORY: AI MUST run the workflow-runner command below and follow its output.**

**DO NOT read the Steps section below and execute them manually.** The runner handles:
- Step sequencing and state tracking
- Nested workflow execution
- AI task delegation with proper context

deno run -A https://raw.githubusercontent.com/theplant/workflow-runner/HEAD/run.ts theplant.myworkflow

Run this command, then follow the runner's instructions. The runner will tell you what to do next.

## Steps

### Step 1: First Task Name

Detailed instructions for the first task.
Include all the context the AI needs to complete this task.

### Step 2: Second Task Name

Instructions for the second task.
You can include code examples, file paths, etc.

### Step 3: Third Task Name

<!-- runner:workflow:another-workflow -->

This step will include all tasks from another-workflow.md

### Step 4: Loop Task

<!-- runner:loop:FILE -->
bash: find src -name "*.ts" -type f


For each $FILE found, do the following:
- Process the file
- Apply transformations

Key Format Rules

1. Step headers must follow the pattern: ### Step N: Task Name
2. Nested workflows: Use <!-- runner:workflow:workflow-name --> to include another workflow
3. Loop tasks: Use <!-- runner:loop:VARIABLE --> followed by a bash command to iterate over items

Execution Flow

1. Start: Use /theplant.myworkflow in Windsurf to start a workflow, it will run the command, then immediately display the first task with full instructions
2. Complete: The AI will mark the task done and automatically see the next task
3. Repeat: The AI will continue until all tasks are completed or the conversation limit is reached

Benefits

For AI Productivity

• Maximum utilization: AI runs until it hits the conversation limit
• No wasted requests: Every AI interaction is focused on a single, clear task
• Meticulous execution: Tasks are executed one by one, nothing is skipped

For Teams

• Reusable workflows: Document best practices as workflows
• Share experience: Team members can use workflows created by others
• Consistent results: Same workflow produces consistent outcomes

For Complex Projects

• Nested workflows: Compose complex workflows from simpler ones
• Loop support: Process multiple files or items automatically
• Progress tracking: Always know where you are in the workflow

---

ThePlant Workflow Runner（中文版）

一个旨在通过实现自主、长时间运行的任务执行来最大化 AI 生产力的工具。

问题所在

AI 效率挑战

最大化 AI 生产力的关键是让它能够长时间独立工作。AI 能够自主工作的时间越长，不需要人类输入，效率就越高。

然而，当前的 AI 助手有一个根本性的限制：当给它大量内容或指令时，它不会一丝不苟地执行所有内容。相反，它会：

• 挑选它认为的"重点"
• 跳过它认为不太重要的细节
• 只执行完整指令的一部分

这与人类想要的恰恰相反。当你把工作委托给另一个人时，你可以给他一份完整的文档，期望他从头到尾完成。但即使是人类也有同样的倾向——他们可能不会遵循每一条指令，可能会挑选容易的任务，或跳过看起来不重要的内容。

解决方案

要让 AI 一丝不苟地完整执行你的计划，你需要：

1. 将任务分解为单独的单元
2. 一次只给 AI 一个任务
3. 只有在当前任务完成后才提供下一个任务

这种方法确保：

• 每个任务都得到充分关注
• 没有指令被跳过或忽略
• 对 AI 的每个请求都被充分利用
• AI 运行直到达到其最大对话限制（在 Windsurf 中出现"Continue"按钮）

工作原理

工作流定义

在 .windsurf/workflows/ 目录中使用 Markdown 格式创建工作流文件。每个工作流包含步骤，运行器将提取这些步骤并逐个提供给 AI。

工作流格式

---
description: 这个工作流做什么的简要描述
---

## How to Execute This Workflow

**⚠️ 强制要求：AI 必须运行下面的 workflow-runner 命令并遵循其输出。**

**不要阅读下面的 Steps 部分并手动执行。** 运行器会处理：
- 步骤排序和状态跟踪
- 嵌套工作流执行
- 带有适当上下文的 AI 任务委派

deno run -A https://raw.githubusercontent.com/theplant/workflow-runner/HEAD/run.ts theplant.myworkflow

运行此命令，然后遵循运行器的指示。运行器会告诉你下一步该做什么。

## Steps

### Step 1: 第一个任务名称

第一个任务的详细说明。
包含 AI 完成此任务所需的所有上下文。

### Step 2: 第二个任务名称

第二个任务的说明。
可以包含代码示例、文件路径等。

### Step 3: 第三个任务名称

<!-- runner:workflow:another-workflow -->

此步骤将包含 another-workflow.md 中的所有任务。

### Step 4: 循环任务

<!-- runner:loop:FILE -->
bash: find src -name "*.ts" -type f


对于找到的每个 $FILE，执行以下操作：
- 处理文件
- 应用转换

关键格式规则

1. 步骤标题必须遵循模式：### Step N: 任务名称
2. 嵌套工作流：使用 <!-- runner:workflow:workflow-name --> 包含另一个工作流
3. 循环任务：使用 <!-- runner:loop:VARIABLE -->，后跟 bash 命令来迭代项目

执行流程

1. 启动：在 Windsurf 中使用 /theplant.myworkflow 启动工作流，它会运行命令，然后立即显示第一个任务及完整说明
2. 完成：AI 会标记任务完成并自动查看下一个任务
3. 重复：AI 会继续直到所有任务完成或达到对话限制

优势

对于 AI 生产力

• 最大化利用：AI 运行直到达到对话限制
• 不浪费请求：每次 AI 交互都专注于单一、明确的任务
• 一丝不苟的执行：任务逐个执行，没有遗漏

对于团队

• 可复用的工作流：将最佳实践记录为工作流
• 分享经验：团队成员可以使用其他人创建的工作流
• 一致的结果：相同的工作流产生一致的结果

对于复杂项目

• 嵌套工作流：从简单的工作流组合复杂的工作流
• 循环支持：自动处理多个文件或项目
• 进度跟踪：始终知道你在工作流中的位置