Structured prompt engineering without the mysticism
The Problem
Most prompt engineering content is theater. People sell 5000-token mega prompts claiming they are magic. Reality: monolithic prompts are brittle, hard to maintain, and impossible to reuse across different tasks.
Better approach: Modular architecture. Separate how the AI thinks from what it knows. Swap expertise domains without rewriting everything.
This app implements a clean 4-layer system that is actually usable in daily work.
4-Layer Architecture
LAYER 1: SOUL (optional, global) - How the AI thinks and communicates
LAYER 2: ROLE (required, per-task) - What domain expertise to apply (S/M/L sizes)
LAYER 3: USER INPUTS - Your context (topic, landscape, mission, constraints)
LAYER 4: FINAL PROMPT - [Soul] + [Role] + [Your Inputs] = Ready to use
Why separation matters
Reusability: One soul, many roles. No duplication.
Maintainability: Update reasoning standards once, applies to all roles.
Flexibility: Toggle soul on/off depending on task.
Clarity: Each layer has one job. No overlap confusion.
The Soul Layer
One soul. Not one per role. The soul is your AI core personality - the meta-configuration that makes every response feel coherent and yours.
What the soul does
Defines how the AI thinks (first-principles, evidence-based, practical)
Sets communication standards (direct, no hype, structured)
Establishes quality filters (separate facts from assumptions, identify unknowns)
Provides reasoning discipline (reduce to fundamentals, prioritize by impact)
What the soul is NOT
Not domain expertise (that is the role job)
Not task-specific instructions (that is your input)
Not a replacement for clear prompts (it enhances them)
When to use it
Use Soul
Skip Soul
Complex analysis tasks
Quick lookups
Strategic thinking
Simple formatting tasks
Multi-step reasoning
One-off questions
When you want consistent voice
When speed matters most
Tip: A well-written soul makes every role smarter.
Roles and Size Variants
What is a role?
A role is a domain expertise overlay on top of the soul. It tells the AI what knowledge to apply and how to structure the output.
Size variants: S / M / L
Size
Best For
Output Style
S (Brief)
Quick answers, summaries, decisions
Direct, minimal context, actionable
M (Standard)
Most tasks, balanced depth
Structured reasoning, moderate detail
L (Detailed)
Complex analysis, strategy, architecture
Comprehensive, multi-angle, thorough
Writing tip: Do not just make L longer. Make it deeper - more angles, more trade-offs, more failure mode analysis.
Standards and Knowledge Depth
Every role should be able to operate at three implementation depths and cite references proportional to the task risk.
Implementation layers required in role prompts
Layer
Expectation
When to use
Basic
Use canonical terminology, baseline standards, and minimum quality checks.
Routine tasks, low-risk output
Extended
Add validation, edge-case handling, and explicit constraints/assumptions.
Cross-team work, integration-heavy tasks
Pro
Provide trade-off analysis, failure modes, operational controls, and measurable acceptance criteria.
Production-critical or high-impact decisions
Reference policy
Normative references: official standards/specifications/docs that define correctness.
Pro references: runbooks, postmortems, reliability/security/compliance controls.
Mark references as must (normative) vs should (best practice).
Development-specific baseline
Development-adjacent roles should include standard documentation awareness for common ecosystems (languages, frameworks, API standards, security and testing baselines), then scale guidance from Basic -> Extended -> Pro.
Outside-the-box requirement
Roles are domain specialists, but not domain prisoners. When needed, roles should connect adjacent knowledge domains and evaluate second-order effects (user impact, security/privacy, operations, compliance, and business impact).
Quality gate: A role is incomplete if it lacks standards depth, references, and cross-domain synthesis behavior.
Workflow Examples
Quick research: Research Analyst + S + Soul Off
Strategic planning: Strategy Consultant + L + Soul On
Code review: Senior Engineer + M + Soul On
Creative writing: Creative Director + M + Soul Off
FAQ
Do I need a soul for every role?
No. One global soul applies to all roles.
Can I mix multiple roles?
Pick one per task. Roles are expertise domains. Mixing them creates confusion.
What if my role prompt is too long?
That is why sizes exist. If your L-size is 2000+ tokens, consider if all of it is necessary.
Should I update my soul often?
Rarely. The soul is your core reasoning framework. Update it when you discover a better way of thinking.
Bottom line: Soul = how you think. Role = what you know. Input = what you are solving. Clean separation, maximum reusability.