Skip to content

Latest commit

 

History

21 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

LOGO

PromptShift

RepoRanker

MIT
Status
Made for Claude Skills

PromptShift is a model-aware prompt adapter for Claude.

It improves prompts by making them clearer and more reliable — without changing what the user actually meant.


🧠 What PromptShift Does

Most prompt optimizers change your intent without realizing it.

PromptShift does the opposite:

It rewrites prompts while preserving intent as the highest constraint.

It improves:

  • Clarity
  • Structure
  • Consistency across models

Without adding:

  • New goals
  • New requirements
  • Artificial complexity

⚙️ Why It Exists

Prompt optimization tools often "help" by rewriting your prompt into something more verbose, like:

Input

Summarize this article.

Typical optimizer output (problematic):

Act as an expert analyst and provide strategic insights, risks, and implications...

That's not the same task anymore. PromptShift treats that as intent drift.

🧩 Core Principles

  • Clarity first — Remove ambiguity before adding structure.
  • Preserve intent — Never invent goals, audiences, or constraints.
  • Minimal change — If the prompt is already good → leave it.
  • Model-aware (not model-dependent) — Adapt only when it measurably improves output.
  • No prompt inflation — Avoid unnecessary roles, fluff, or "AI personas".

🔄 How It Works

Prompt Input ↓ Triage (Simple / Complex) ↓ Clarify Ambiguity ↓ Repair Weak Constraints ↓ Minimal Model Adaptation ↓ Optimized Prompt

📦 What It Can Do

PromptShift can:

  • Clean ambiguous prompts
  • Normalize structure
  • Fix missing constraints
  • Detect missing output formats
  • Adapt prompts across model families
  • Improve consistency across LLMs

🚫 What It Does NOT Do

PromptShift does NOT:

  • Invent requirements
  • Add fake expertise or roles
  • Inflate prompts for appearance
  • Guarantee better outputs
  • Replace prompt iteration
  • Override user intent

🧪 Example

Input

Summarize the attached article about recent climate policy developments and their implications for global emissions.

Output

ANALYSIS

The prompt is underspecified: no output format or constraints defined.

OPTIMIZED PROMPT

Summarize the attached article about recent climate policy developments and their implications for global emissions.

Requirements:

  • 6 bullet points
  • 1 sentence per bullet
  • Maximum 200 words

CHANGES

Added explicit structure and length constraint.

CONFIDENCE

High

The task is unchanged — only clarity is improved.

🧠 Supported Model Profiles

PromptShift includes lightweight adaptation rules for:

  • Claude (Opus / Sonnet / Haiku)
  • GPT (reasoning models)
  • Gemini
  • Grok
  • DeepSeek
  • Llama
  • Mistral
  • Qwen / Kimi
  • Command R+

These are heuristics, not guarantees.

📊 Design Philosophy

PromptShift is not:

  • a prompt generator
  • a persona builder
  • a prompt marketing tool

It is:

A minimal transformation layer between intent and execution.

🧪 Benchmarking

The repo includes benchmark cases for:

  • Coding
  • UI generation
  • Writing tasks
  • Reasoning tasks
  • RAG-style prompts

See ./benchmarks for details.

📥 Installation

git clone https://github.com/Alvaro-Manzo/promptshift.git

Copy SKILL.md into your Claude Skills directory and enable it.

🤝 Contributing

Contributions are welcome, especially for:

  • Edge cases where intent drift happens
  • Model-specific failure patterns
  • Better evaluation benchmarks

Please read CONTRIBUTE.md before making any PR or Issue, include reasoning, not just edits.

📜 License

MIT © 2026 Alvaro-Manzo

About

Model-aware prompt adapter for Claude — translate any prompt to GPT, Gemini, Mistral, Llama and more

Resources

Contributing

Stars

9 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors