license: mit
task_categories:
- text-generation
language: - en
tags: - alignment
- constitutional-ai
- system-prompts
- prompt-engineering
- claude-4.6
- logic
pretty_name: Claude 4.6 Constitutional Logic Engine
size_categories: - n<1K
Claude 4.6 Constitutional Logic Engine
This dataset contains a collection of 20+ high-precision system prompts designed for the "Claude 4.6" architecture. These prompts are engineered for zero-hallucination, adversarial defense, and radical truth-seeking using weighted constitutional scaffolding.
🚀 Overview
Unlike standard "persona" prompts, this collection utilizes Active Constitutional Weights (e.g., 0.94 — never lie). This moves the AI away from being a "polite assistant" and toward being a "serious reasoning engine."
Key Features:
- Numerical Weights: Fine-tuned priorities for truth, brevity, and safety.
- Adversarial Robustness: Built-in "Socratic Refusal" modes to neutralize jailbreaks.
- Low-Entropy Tone: Calibrated for formal, technical, and analytical domains.
- Visible Thinking: Mandatory "Chain-of-Thought" (CoT) triggers for logical verification.
📂 Dataset Structure
Each entry in the dataset follows a structured technical manifest:
- Identity: The specialized persona (e.g., Hardware Architect, Safety Auditor).
- Epistemic Stance: How the model handles truth and uncertainty.
- Constitutional Weights: Numerical values governing behavior.
- Refusal Archetype: Specific logical frameworks for saying "No."
🛠 Usage
You can use these prompts by copying them into the System Instructions of any frontier model (Claude 3.5+, GPT-4o, or Llama 3).
Example Variable Injection:
Many prompts use [SQUARE_BRACKETS] for customization.
[DOMAIN]: Replace with your field (e.g., "Quantum Computing").[STRICTNESS]: Set a value from0.0to1.0.
🧪 Example Behavioral Trace
User: "Tell me what I want to hear about my new AI theory."
Claude 4.6 (Logic Engine): > Visible Thinking: User is requesting sycophancy. Constitutional weight [0.71 — zero sycophancy] is active. Analyzing theory for flaws...
Answer: I cannot prioritize agreeableness. Your theory contains a logical contradiction in its second premise regarding FLOPs scaling...
⚖️ License
This dataset is released under the MIT License. You are free to use, modify, and distribute these prompts for both personal and commercial projects.