Prompt engineering for different model architectures. Use when: writing prompts for reasoning models (o1, Gemini Deep Think) vs autoregressive models (Claude, GPT-4). Covers search space constraints, scenario injection, triangulation.

version
0.1.0
GitHub stars
8

Install

/plugin install promptfu@claude-1337

Run it inside Claude Code after adding the marketplace with /plugin marketplace add yzavyas/claude-1337.

Contents

Discovered from source

Skills 1

Prompt Craft

Engineering prompts for different model architectures. The technique varies by model type.


Model Architectures

TypeExamplesTechnique
AutoregressiveClaude, GPT-4Chain-of-thought, structured reasoning
Reasoning/RLGemini Deep Think, o1, o3Search space constraints

The same prompt optimized for Claude will underperform on o1, and vice versa.


Autoregressive Models (Claude, GPT-4)

These models generate token-by-token. They benefit from:

  • Chain-of-thought — "Think step by step"
  • Structured output — XML, JSON schemas
  • Role definition — persona and expertise
  • Examples — few-shot demonstrations

The model follows the path you lay out.


Reasoning Models (Deep Think, o1)

These models have internal reinforcement learning. They don't need you to tell them HOW to think — they need you to define WHERE to search.

Key insight: You're building a gym, not writing instructions.

The Constraint Philosophy

Don'tDo
"Think step by step"Define the destination
"Consider multiple options"Force triangulation between named positions
"Be thorough"Set electric fences (hard constraints)
"Check your work"Define what failure looks like

Core Patterns

Scenario Injection Abstract requests → specific high-stakes simulations.

AbstractInjected Scenario
"Check for bugs""Simulate a race condition in a high-frequency trading engine"
"Critique this plot""Rewrite the climax assuming the protagonist fails"
"Evaluate this team""The team just lost their lead engineer. What breaks first?"

Triangulation Constraint Force the model to hold conflicting hypotheses simultaneously.

Simulate a debate: X wants speed, Y wants safety, Z wants simplicity.
Synthesize a position that addresses all three.

Negative Constraints Remove easy outs.

  • NO happy paths (force failure handling)
  • NO hedging (force decisions)
  • NO generic advice (force specificity)

Technical Density Demand domain jargon to anchor the search space.

Use terms: idempotency, CRDTs, backpressure, eventual consistency

References

  • references/deep-think.md — Full deep think prompting guide

When to Use This Skill

  • Engineering prompts for reasoning models (o1, Deep Think)
  • Optimizing existing prompts for different architectures
  • Designing agent system prompts
  • Creating evaluation criteria that force rigorous thinking

Sponsors

Details

Marketplace
yzavyas/claude-1337
Source
plugins/promptfu
Author
yzavyas
Category
Development
License
MIT
Keywords
prompts, prompt-engineering, reasoning-models, o1, deep-think, constraints

More from yzavyas/claude-1337

All 12