Prompt Architect — 27 Frameworks for Expert Prompts

What it is

Prompt Architect is a Claude Code skill that systematically transforms vague prompts into structured, expert-level prompts using 27 research-backed frameworks organized across 7 intent categories. It scores every prompt on 5 dimensions: clarity, specificity, context, completeness, and structure, then recommends and applies the optimal framework.

Prompt Architect is designed for developers and prompt engineers who want consistent, high-quality prompts across AI coding tools like Claude Code, Cursor, ChatGPT, and Gemini CLI.

How it saves time or tokens

Manually optimizing prompts requires trial and error across multiple iterations. Prompt Architect eliminates this by analyzing your intent and applying a proven framework in one step. Better-structured prompts produce more accurate outputs on the first attempt, reducing the back-and-forth that wastes tokens. The 5-dimension scoring gives immediate feedback on prompt quality before sending to the model.

How to use

1. Install via npx:

npx @ckelsoe/prompt-architect

1. The interactive installer auto-detects your AI agents (Claude Code, Cursor, etc.).

1. Select installation targets and the skill is ready to use.

1. In your AI tool, describe any task and the skill analyzes, scores, and optimizes your prompt automatically.

Example

Using the CO-STAR framework for content creation:

Original prompt:
'Write a blog post about RAG'

Optimized with CO-STAR:
Context: Technical blog for AI engineers building production systems
Objective: Explain RAG architecture with practical implementation steps
Style: Tutorial with code examples, authoritative but accessible
Tone: Professional, direct, no marketing language
Audience: Mid-level developers familiar with Python and LLMs
Response: 1500-word article with architecture diagram and code samples

The RISEN framework for research tasks:

Role: Senior ML engineer reviewing embedding strategies
Instructions: Compare dense vs sparse embeddings for e-commerce search
Steps: 1) Define metrics 2) Benchmark on sample data 3) Recommend
End goal: Decision document with quantitative comparison
Narrowing: Focus on latency under 100ms, catalog size 1M+ items

Related on TokRepo

• Prompt library — curated prompts for various AI tools
• Coding tools — AI-assisted development tooling

Common pitfalls

• Installing without a GitHub token with read:packages scope causes npm authentication errors
• Applying high-complexity frameworks like CO-STAR to simple one-shot questions adds unnecessary overhead
• The skill requires .npmrc configuration for the GitHub package registry; ensure this is set before running npx