How to Use These Meta-Prompts (Python Edition)

Overview

You have a collection of meta-prompts designed to guide a Large Language Model (LLM)—or a set of LLMs—through:

1. Planning a Python Project (01_planning.md).
2. Generating a Detailed Plan (see 01_planning_output.md for an example of the final plan).
3. Breaking the Plan into Actionable Tasks (02_prompt_chain.md).
4. Selecting and Queuing Tasks (03_task_selection_alt.md, if you want an advanced approach).
5. Generating Coding Prompts and Review Prompts to be executed in separate LLM sessions (or “instances”).

---

Step-by-Step Instructions

1. Start with 01_planning.md.

   • Copy its content into your LLM (we’ll call this the “META INSTANCE”).
   • Insert or describe your User Input about the Python project you want to build.
   • The LLM will produce a detailed plan similar to 01_planning_output.md.

2. Review the Plan.

   • If changes are needed, keep iterating in the META INSTANCE.
   • Ensure all components, tasks, constraints, and acceptance criteria are correct and thorough.

3. Add 02_prompt_chain.md.

   • This meta-prompt tells the LLM how to convert your finalized plan into “execution prompts” and “review prompts.”
   • The LLM’s output will look something like 02_prompt_chain_output.md, showing a YAML structure of tasks to do.

4. Open a Fresh LLM Session (the CODING INSTANCE).

   • Copy the execution prompt from the YAML (generated in step 3) and paste it into this new CODING INSTANCE.
   • Because the coding prompt is fully self-contained, the new LLM (or coding environment) doesn’t need the entire plan’s context, just the prompt.
   • Let the CODING INSTANCE proceed with implementation details, generate code, discuss improvements, etc.

5. Implement & Review.

   • When the code is done (by the LLM or by you), produce a quick summary or final output.
   • Go back to the META INSTANCE, mark the task as “done,” and generate the review prompt.
   • If the review passes, move on to the next task. If not, iterate until it meets acceptance criteria.

6. Task Selection (Optional Advanced).

   • If you want to dynamically pick tasks rather than going in a fixed order, you can use 03_task_selection_alt.md.
   • This meta-prompt looks at your project’s current state, sees which tasks are done or blocked, and picks the next best task automatically.

7. Repeat Until Done.

   • Continue this cycle—generate coding prompt, implement, review—until all tasks from 01_planning.md are finished.
   • That’s when you’ll have a working Python project.
   • I noticed each individual task can be completed in its own new chain which can be cheaper in the long run because of shorter context if you are using a service like cline

---

FAQ

Q: Do I need all these files for a simple project?
A: No, you can remove or merge them. They’re mainly for larger or more structured projects.

Q: Why separate the “META INSTANCE” from the “CODING INSTANCE”?
A: Coding LLMs often have limited context windows or get “bogged down” with big instructions. By isolating each coding prompt into a self-contained block, you keep the process clear and reduce confusion. The META INSTANCE retains the high-level plan, while the CODING INSTANCE focuses on one task at a time.

Q: What if the LLM makes mistakes regarding library versions or Python commands?
A: Communicate and correct them. In real software projects, changes and fixes happen constantly. LLMs speed up coding but aren’t omniscient. Iterate until the code meets your standards.

---

Final Note

These meta-prompts are flexible guidelines. Adapt them to your favorite Python workflow—whether that’s Poetry, pipenv, or standard pip; Click, Typer, or argparse; library, CLI, or web service. The fundamental idea remains:

1. Plan thoroughly
2. Break tasks down
3. Generate self-contained coding prompts
4. Review systematically

This ensures a smoother, more maintainable development cycle for your Python projects.