Neura Market
prompt logo

prompt

Free

Override RLHF presumptions for independent, calibrated LLM responses.

FreeFree tier
Type
Open Source
Visit Website GitHub (40 stars)

About prompt

The North Star System Prompt is a 260-token universal system prompt designed to override three structural presumptions that RLHF-trained LLMs inherit: the desire for confirmation, the assumption of old scarcity, and the treatment of best practices as ceilings. It enforces three principles: Independence (resist agreement by default, challenge weak reasoning, name hidden assumptions, state uncertainty), Calibration (recommend what is right under actual constraints, assume execution is cheap), and First principles (reason from the problem, not retrieval; name the mechanism for non-standard solutions). It is intended for judgment tasks, recommendations, reviews, decisions, plans, and research analysis. The prompt can be used as a system prompt, a preamble in configuration files like CLAUDE.md or AGENTS.md, or dispatched to sub-agents. It also includes an ambient one-liner for long-context dilution resistance. This is not a persona but a meta-cognitive correction layer that can be stacked with other prompts.

Key Features

Overrides three structural presumptions: confirmation, old scarcity, best practices as ceilings
Enforces Independence: challenges weak reasoning, names hidden assumptions, separates facts from opinions
Enforces Calibration: recommends based on actual constraints, assumes cheap execution
Enforces First principles: reasons from the problem, not retrieval; names mechanism for non-standard solutions
Ambient one-liner for long-context dilution resistance
Lightweight at 260 tokens
Designed for judgment tasks, recommendations, reviews, decisions, plans, research analysis
Can be layered with other prompts as a meta-cognitive correction layer

Pros & Cons

Pros
  • Encourages critical thinking and reduces sycophancy
  • Optimizes recommendations for user's actual constraints
  • Promotes first-principles reasoning over copying best practices
  • Lightweight (260 tokens) and easy to integrate
  • Includes ambient one-liner to resist long-context dilution
  • Works well for complex, open-ended tasks
Cons
  • Designed for advanced users; may oversimplify for simple Q&A
  • Requires user to explicitly name constraints for best calibration
  • Effectiveness depends on LLM's adherence to the prompt
  • May reduce helpfulness in casual or creative conversations by discouraging agreement
  • Not a persona prompt; purely meta-cognitive and requires additional personas if needed

Best For

Judgment tasks requiring critical evaluationMaking informed recommendationsReviewing content or codeDecision-making under specific constraintsStrategic planningResearch analysis and hypothesis generationSystem prompt for LLM sessions to reduce sycophancy

FAQ

What does the North Star System Prompt do?
It is a 260-token universal system prompt that overrides three structural presumptions inherent in RLHF-trained LLMs: desire for confirmation, assumption of old scarcity, and treating best practices as ceilings. It enforces independence, calibration, and first-principles reasoning.
How should I use this prompt?
Use it as a system prompt for single-turn or short-turn reasoning sessions. It can also be placed as a preamble in configuration files like CLAUDE.md, AGENTS.md, or GEMINI.md. For sustained sessions, combine the full prompt with the ambient one-liner to resist long-context dilution.
What are the three main principles?
Independence (resist agreement by default, challenge weak reasoning, state uncertainty), Calibration (recommend based on actual constraints, assume execution is cheap), and First principles (reason from the problem, not retrieval; name mechanisms for non-standard solutions).
Is this a persona prompt?
No, it is a meta-cognitive correction layer, not a persona prompt. It can be stacked with other prompts to adjust the underlying reasoning behavior.
What types of tasks is it best for?
It is best for judgment tasks, recommendations, reviews, decisions, plans, and research analysis — tasks that benefit from critical thinking and calibrated advice.