DEV Community

Ted Enjtorian
Ted Enjtorian

Posted on

[POG-Task-05] Treat Tasks and Prompts as Code

#promptengineering #softwareengineering #ai #productivity

AI Native Task Governance Model: Making AI collaboration painless and enabling experience accumulation!

What Was Valuable in the Past? What's Most Expensive Now?

Think back—what did we consider most valuable in the past?

  • Beautifully written and detailed documentation
  • Smoothly running, bug-free codebase
  • Small tools that saved us tons of time

Yes, these are all great. But entering this era of working alongside AI, a new gold standard has emerged: The "Tasks" you define and the "Prompts" you write are extremely valuable!

Unfortunately, many haven't realized this yet. We often treat our communication with AI like disposable chopsticks. Once the chat window is closed, that perfect, thoughtfully refined piece of Prompt simply vanishes into the sea of history.

Let's Look at Our Current "Ungoverned" Workflow

This is the common daily development routine across many teams today (especially within an IDE x Git collaborative environment):

  1. Requirements come in.
  2. They turn into a Project, then get stuffed into Jira & Excel for tracking.
  3. We proceed with Assignment.
  4. Inside the IDE, developers begin navigating through Source Code, SA Docs, SD Docs, and Analytics trying to find context.
  5. Entering the Execution Phase: The developer inputs a Prompt, letting the AI Agent help write code and execute the task.
  6. During this process, the AI Agent is quietly recording the following in the background, deep within "hidden chat blocks":
    • Session Context
    • Plan Formulation
    • Walkthrough (Execution Summary)
    • Task Breakdown

Sounds smooth, right? But here is the problem: These precious records meticulously generated by the Agent are buried deep within chat histories, making them extremely difficult to find and retrieve! When the team encounters a similar task next time, they can't locate previous records and are forced to draft a Prompt from scratch and make the AI break down the task all over again. Your distilled "experience" is simply wasted.

Let's Look at Our Current

Enter the Improved Workflow: Introducing POG-Task

If code can be version-controlled and documentation can be preserved, why can't tasks and prompts?
This is the core concept we want to share: "Treat Tasks and Prompts as Code!"

In POG-Task's design philosophy, we give tasks and prompts an entirely new definition:

  • Evolving into "Units of Intention": A task is no longer a fleeting chat message. It becomes a structured, human-and-machine-readable, and Agent-governed explicit intention. This ensures that AI has clear, unambiguous boundaries when executing.
  • Elevated to "First-Class Assets": Your Prompt, task breakdown logic, and the invaluable AI execution history (Record.md) are no longer disposable transition products. They carry the same weight as Source Code—living, Git-versioned assets that accumulate team knowledge.

So, how do we operationalize these philosophies? POG-Task introduces two indispensable "gatekeepers":

  • task.schema.json (The Law of Tasks): This is a JSON Schema defining exactly what a task should look like. It acts as the legal code for your development environment, strictly standardizing task structures. If a generated task doesn't match the format, the system (or AI) simply ignores it, ensuring all requirements stay within a predictable framework.
  • pog-task-agent-instructions.md (Agent Instruction Manual): This is a system protocol written explicitly for AI to read. Before starting any work, the AI must consult this guide to ensure its behavioral standards remain consistent and reliable, preventing it from wildly guessing or diverging.

Let's look at the evolved workflow after introducing POG-Task:

  1. Requirements come in.
  2. First, integrate with Project / Jira & Excel / Assignment.
  3. Next, enter the pog-task governance framework.
  4. This time, when you start working in your IDE, alongside your source code and specs, your project now includes:
    • Task.yaml (Structured Task)
    • docs.md (Comprehensive Task Documentation)
    • Record.md (Invaluable AI Execution Records)
  5. The same AI Agent continues to unleash its power (Context, Plan, Walkthrough, Task).
  6. Finally, execute by inputting the Prompt.

Notice the difference? Through Task.yaml and Record.md, your task breakdown logic and the ultimate Prompt tested through AI interactions are entirely materialized, documented, and codified!

Even better, this improved workflow brings two major breakthroughs:

  1. Transparent and Git-Versioned Agent Records: The context that used to be hidden in conversations is now transformed into traceable, version-controllable physical files. Say goodbye to lost history.
  2. Distilling and Reusing Skills: You can re-analyze Record.md at any time to extract the AI's problem-solving methods and experience into reusable "Skills", continuously evolving your entire team's capabilities!

Enter the Improved Workflow: Introducing POG-Task

Preserve Your Personal Value, Starting Now!

Every Prompt you create is an intangible asset. To help you seamlessly transition into this workflow, we have developed a dedicated developer tool:

Download the VSCode Plugin Now: POG Task Manager
Search for it in the marketplace to explicitly turbocharge your AI collaboration with these three core features:

  1. Visual Task List Management Use an intuitive tree view in the sidebar to grasp all structured tasks. It actively monitors file changes in real-time and supports status filtering, giving your team perfect visibility over all work.
  2. Built-in Agent Prompt Templates Provides standardized dialogue instructions (e.g., creating a new task, executing a task). It automatically injects project context and supports "one-click copy," saving you from writing lengthy Prompts from scratch.
  3. Task Detail & Reasoning Record Inspection Review or edit task details effortlessly via a user-friendly Webview interface. With a single click, open the corresponding Record.md to instantly trace back the AI Agent's thought processes and execution context.

Initialize Immediately (Init): After installation, the extension will automatically detect your project status and proactively prompt you. One click, and your workspace is fully initialized in seconds!

Stop letting your hard-earned wisdom disappear into the endless river of chat histories. Starting today, manage and reuse your Tasks and Prompts just like you do your Code!

Full Document : https://enjtorian.github.io/pog-task/

Get Pog Task Manager : https://marketplace.visualstudio.com/items?itemName=enjtorian.pog-task-manager

POG Task Manager

Top comments (0)