🚀 LLM Prompting Master Template
Based on Anthropic's Expert Workflows
Use this document as a guide to structure your interactions with LLMs. For complex problems, move through these stages sequentially.
1. The Context Layer (Memory Injection)
Goal: Establish a permanent "vibe" and technical baseline so you don't repeat yourself.
- When to use: At the very start of a new project or chat thread.
- Template:
"You are my [Role]. Remember these preferences: [Preference 1], [Preference 2], and [Preference 3]. Acknowledge these and apply them to all future responses in this thread."
2. The Discovery Phase (Reverse Prompting)
Goal: Prevent the AI from guessing. Force it to audit your requirements first.
- When to use: Before you let the AI start a complex task (coding, strategy, writing).
- Template:
"I need to [Goal]. Before you begin, ask me [Number] clarifying questions about my [Dataset/Context/Outcome]. Do not start the task until I have answered these questions."
3. The Multi-Expert Audit (Role Stacking)
Goal: Eliminate bias and catch blind spots by simulating a panel of experts.
- When to use: High-stakes decision making or analyzing a strategy.
- Template:
"Analyze [Topic/Problem] from three perspectives simultaneously:
[Expert Role 1 - e.g., Critical Skeptic]
[Expert Role 2 - e.g., Creative Optimist]
[Expert Role 3 - e.g., Practical Implementation Lead] Show all three viewpoints and where they conflict."
4. The Execution Chain (Constraint Cascade)
Goal: Prevent "model fatigue" by breaking the task into logical, progressive steps.
- When to use: For long-form content, complex code, or deep research.
- Template:
Step 1: "First, [Simple Task]. Wait for my response."
Step 2: "Now that you've done that, [Add Complexity/Layer]."
Step 3: "Finally, [Final Polish/Output Format]."
5. The Quality Gate (Verification Loop)
Goal: Self-correction. Force the AI to find its own mistakes before you have to.
- When to use: For anything that requires high accuracy (math, code, logic).
- Template:
"Complete [Task]. Once finished, identify 3 potential bugs, logical fallacies, or edge cases in your own work. Then, provide an updated version that resolves those issues."
Summary Table for Quick Reference
| Technique | Core Benefit | Best For... |
|---|---|---|
| Memory Injection | Consistency | Coding & Style |
| Reverse Prompting | Accuracy | Project Planning |
| Role Stacking | Depth | Strategy & Reviews |
| Constraint Cascade | Logic | Complex Workflows |
| Verification Loop | Reliability | Debugging & Fact-checking |