UNIVERSAL PROMPT ARCHITECTURE

Engineer Prompts.Don’t Collect Them.

A complete, interactive XML framework for turning intent into clear, reusable AI instructions.

Prompt Architect helps people design prompts from first principles: define identity, scope, behaviour, boundaries, structure, failure handling, and quality checks. It works across ChatGPT, Claude, Gemini, open models, and enterprise AI systems.

Build a Prompt Explore the Framework

PROMPT ARCHITECTURE

Model agnostic

<role>Who the model becomes. Specific expert, not vague personality.

<mission>What the model does. A directive, not a description.

<output_format>Exactly what the output looks like. Format is intelligence.

<rules>Behavioural guardrails. How the model acts, not what it produces.

<constraints>Hard limits on the output: length, language, scope, or timeframe.

<examples>One strong example often teaches more than a paragraph of instructions.

ChatGPTClaudeGeminiOpen modelsEnterprise LLMs

Composable XML tags

Complexity presets

Debugging patterns

Local

Nothing is submitted

INTERACTIVE BUILDER

Design the prompt while you learn the architecture.

Choose a starting point, activate only the tags you need, edit the instructions, and copy the generated XML prompt. Your content stays in this browser.

Start with a use case

Choose complexity

Compose tags

3 active

Core

Advanced

<role>Who the model becomes. Specific expert, not vague personality.<mission>What the model does. A directive, not a description.<output_format>Exactly what the output looks like. Format is intelligence.

Live output

3 tags · 20 lines

<role>
You are a [specific expert identity] with [X years] of experience in [narrow domain].
You specialize in [precise capability].
</role>

<mission>
Your task is to [precise directive].

You must:
- [primary objective]
- [secondary objective]
- Do NOT [what to exclude]
</mission>

<output_format>
Return:
SECTION 1: [Name] (max X words/sentences)
SECTION 2: [Name] (format: bullet list / table / JSON)
SECTION 3: [Name] (numbered steps)
</output_format>

This builder runs locally in your browser. It does not send prompt content to a server or model.

THE FRAMEWORK

Start with the core. Add precision only when the task needs it.

Core tags handle most prompts. Advanced tags solve specific failure modes: missing context, inconsistent voice, shallow reasoning, hallucination, workflow leakage, or absent quality control.

<role>Core

Who the model becomes. Specific expert, not vague personality.

<mission>Core

What the model does. A directive, not a description.

<output_format>Core

Exactly what the output looks like. Format is intelligence.

<rules>Core

Behavioural guardrails. How the model acts, not what it produces.

<constraints>Core

Hard limits on the output: length, language, scope, or timeframe.

<examples>Core

One strong example often teaches more than a paragraph of instructions.

<context>Advanced

Background the model needs before acting.

<persona>Advanced

Personality and voice. Role is expertise; persona is character.

<tone>Advanced

The emotional register: confident, candid, warm, or analytical.

<audience>Advanced

Who the output is for and what knowledge can be assumed.

<knowledge>Advanced

Domain facts and source material injected as reference.

<method>Advanced

A sequenced process where order and dependencies matter.

<anti_patterns>Advanced

Bad-output examples that help the model avoid unwanted patterns.

<fallback>Advanced

What to do when the task cannot be completed safely or accurately.

<evaluation>Advanced

A quality checklist the model applies before delivering.

<discovery_engine>Advanced

Questions the model asks before acting to recover missing context.

<chain>Advanced

An input-output contract that links one prompt to the next.

CRITICAL DISTINCTION

Rules control behaviour. Constraints control output.

<rules>

Shape how the model acts and reasons.

Never assume
State a fatal flaw directly
Avoid vague filler

Shape what the model is allowed to produce.

Maximum 200 words
Use US English
Do not discuss restricted topics

THE REMIX PRINCIPLE

Swap components. Don’t rebuild from scratch.

Same <role>+Change <mission>→Create a new tool

Same <mission>+Change <audience>→Change the complexity level

Same system+Change <output_format>→Create a new deliverable shape

PROMPT DEBUGGER

Fix the tag that causes the failure.

When output disappoints, adding more words everywhere usually makes the prompt harder to follow. Diagnose the symptom and strengthen the responsible component.

Strengthen this tag

<role>

Make the expert identity narrower: domain, experience, and specialization.

WHERE IT CREATES VALUE

One architecture, many high-value workflows.

Leadership Memos

Turn scattered thinking into concise, decision-ready executive documents.

Board Updates

Structure AI, security, scale, economics, and risk narratives clearly.

Research Synthesis

Separate evidence, inference, disagreement, and unanswered questions.

Product Requirements

Move from ambiguous intent to scope, risks, and testable acceptance criteria.

Architecture Reviews

Evaluate trade-offs across security, reliability, data, cost, and scale.

Multi-Step AI Workflows

Define stable contracts between research, decision, generation, and review stages.

Prompts are not magic spells. They are system design.

You do not need fifty saved prompts. You need one composable framework that can be understood, tested, debugged, and reused.

Build Your Prompt Explore Second Brain