Skip to main content

​
Writing a PRD

A PRD (Product Requirements Document) is the starting point of every task flow in TaskGraph. It defines what you’re building and why. A clear PRD dramatically improves the quality of your tasks, your model outputs, and your final product — so it’s worth taking the time to get it right.
You don’t need to define your whole app up front. You can write a focused PRD just for the next feature or module you’re working on.
You can start with an empty project or you can start with a feature PRD on an existing project.
You can add and parse multiple PRDs per project using the —append flag. Running --append again on a PRD you edited only generates tasks for the sections you added.

​
What Makes a Good PRD?

  • Clear objective — what’s the outcome or feature?
  • Context — what’s already in place or assumed?
  • Constraints — what limits or requirements need to be respected?
  • Reasoning — why are you building it this way?
The more context you give the model, the better the breakdown and results.

​
Writing a PRD for TaskGraph

Two example PRD templates are available in .taskgraph/templates/:
  • example_prd.md - Simple template for straightforward projects (.md recommended for better editor support)
  • example_prd_rpg.md - Advanced RPG (Repository Planning Graph) template for complex projects with dependencies
Why .md? While both .txt and .md work, Markdown files provide syntax highlighting, proper rendering on GitHub, and better collaboration through formatted documentation.
You can co-write your PRD with an LLM model using the following workflow:
  1. Chat about requirements — explain what you want to build.
  2. Show an example PRD — share the example PRD so the model understands the expected format. The example uses formatting that work well with TaskGraph’s code. Following the example will yield better results.
  3. Iterate and refine — work with the model to shape the draft into a clear and well-structured PRD.
This approach works great in Claude Code, or anywhere you use a chat-based LLM.

​
Choosing Between Templates

Use example_prd.md when:
  • Building straightforward features
  • Working on smaller projects
  • Dependencies are simple and obvious
Use example_prd_rpg.md when:
  • Building complex systems with multiple modules
  • Need explicit dependency management
  • Want structured guidance on architecture decisions
  • Planning a large codebase from scratch
The RPG template teaches you to think about:
  1. Functional decomposition (WHAT the system does)
  2. Structural decomposition (HOW it’s organized in code)
  3. Explicit dependencies (WHAT depends on WHAT)
  4. Topological ordering (build foundation first, then layers)
For complex projects, using the RPG template with a code-context-aware ai agent produces the best results because the AI can understand your existing codebase structure. Learn more about the RPG method →

​
Where to Save Your PRD

Place your PRD file in the .taskgraph/docs folder in your project.
  • You can have multiple PRDs per project.
  • Name your PRDs clearly so they’re easy to reference later.
    • Examples: dashboard_redesign.md, user_onboarding.md
    • Tip: Use .md extension for better editor support and syntax highlighting

​
Parse your PRD into Tasks

This is where the TaskGraph magic begins. In Claude Code, instruct the agent to generate tasks from your PRD:
The agent will execute the following command which you can alternatively paste into the CLI:
This will:
  • Parse your PRD document
  • Generate a structured tasks.json file with tasks, dependencies, priorities, and test strategies
To be asked about gaps in the PRD first, use the /taskgraph:parse-prd plugin command in Claude Code, or add --clarify in the terminal:
It asks up to 5 questions, each with a suggested answer, saves your answers in the PRD under ## Clarifications, and then generates the tasks from it. Now that you have written and parsed a PRD, you are ready to start setting up your tasks.