AI from AI, by AI, and for AI
The Meta Prompting Directive (MPD) is a system-level workflow for operating large language models as machines rather than conversational partners.
MPD replaces ad-hoc, human-written prompts with a structured, closed-loop process in which AI systems are responsible for compiling and executing their own instructions.
This repository defines MPD as an operational pattern, not a philosophy and not a software library.
Most failures in AI outputs are not caused by model limitations. They are caused by human prompting:
- Imprecise language
- Unstated assumptions
- Forgotten constraints
- Conversational or emotional phrasing
Chat-based interaction assumes shared context. Machines do not have that.
MPD removes the human from the critical instruction path and replaces human-authored prompts with AI-compiled directives.
Humans should declare intent.
Machines should generate instructions.
Machines should execute instructions.
MPD enforces this separation explicitly.
To use MPD, the human does not prompt the AI to perform the task.
The human prompts the AI to translate intent into the prompt that will be used.
In MPD:
- Humans declare intent
- AI generates the executable prompt
- AI executes that prompt
This translation step is mandatory and defines the directive.
The Conductor declares intent and success criteria.
Responsibilities:
- State the objective
- Provide relevant context
- Define high-level constraints
Non-responsibilities:
- Writing prompts
- Specifying formatting or reasoning methods
- Managing execution details
Example:
Produce a 1-page executive summary of Q3 sales. Focus on trends, risks, and one recommendation.
The Architect translates intent into a Master Prompt addressed to an AI executor.
Responsibilities:
-
Eliminate ambiguity
-
Make assumptions explicit
-
Infer missing constraints
-
Specify:
- Role
- Objective
- Audience
- Scope
- Output structure
- Reasoning requirements
The Architect does not generate the final output.
Its output is a machine-addressed directive, not human-facing prose.
The Agent executes the Master Prompt verbatim.
Rules:
- No reinterpretation of intent
- No conversational padding
- Follow constraints exactly
- Perform internal reasoning silently unless explicitly instructed otherwise
The Agent produces the final deliverable.
Human Intent
↓
Architect (AI)
→ Master Prompt
↓
Agent (AI)
→ Final Output
Once the Master Prompt is produced, the human is no longer involved in execution.
- AI systems understand their own failure modes better than humans
- AI reliably specifies constraints humans omit
- Explicit directives outperform assumed context
- Role separation prevents prompt drift and scope creep
MPD does not remove humans from the process.
It confines humans to the part they are best at: declaring intent.
When using an AI system that supports system instructions or persistent context:
- Install the Meta Prompting Directive as a system instruction
- Provide intent only
- Allow the AI to compile and execute its own instructions
One-line activation:
Use the Meta Prompting Directive. Act as Architect first, then Agent. Do not merge phases.
- Instructions are compiled, not authored
- Machines write instructions for machines
- Humans do not manage execution
- Determinism is enforced at the instruction layer
If you use or reference this framework in documentation, tooling, publications, or products, please attribute it as:
Meta Prompting Directive (MPD)
Original concept and framework by the author of this repository.
The Meta Prompting Directive (MPD) framework and documentation are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).
Any accompanying code or tooling in this repository is licensed under the MIT License.
This repository documents a reference pattern for AI instruction workflows.
It is intentionally minimal and expected to evolve through clarification, examples, and edge-case analysis rather than feature expansion.