AI Prompt Architect

AI Prompt Architect is a VS Code and Cursor-compatible extension prototype that reviews weak AI coding prompts, scores them, rewrites them with project context, estimates token savings, and records local acceptance analytics.

What It Does

• Scores prompt quality from 0-100.
• Detects missing context, constraints, architecture, output format, testing, security, and deployment expectations.
• Enriches prompts using workspace signals such as languages, frameworks, package files, Docker files, database hints, and cloud hints.
• Rewrites prompts using compact specialized templates for prompt engineering, software, AI, testing, and documentation work.
• Selects an appropriate agent role for the task, such as Prompt Architect, Solution Architect, Backend Engineer, Frontend Engineer, DevOps Engineer, Security Engineer, QA Engineer, AI/RAG Engineer, or Technical Writer.
• Estimates token usage before and after optimization, including when a vague prompt expands for clarity.
• Lets the developer choose: use optimized prompt, edit optimized prompt, or use original prompt.
• Provides a pre-send chat panel where prompts are reviewed before being copied into the assistant chat.
• Stores local learning data for accepted, edited, and rejected suggestions.
• Provides a dashboard webview for usage metrics.

Platform Notes

VS Code, Cursor, and similar editors do not expose a universal API that can intercept every prompt before it reaches Copilot, Codex, Claude Code, or another assistant. This extension implements the safe, supported pattern: select or enter a prompt, run Prompt Architect: Optimize Prompt, then paste the result into the target assistant.

For a closer pre-send workflow, run Prompt Architect: Open Pre-Send Chat. Type your chat there, review the optimized prompt, then choose Use Optimized Prompt or Ignore Suggestion and Use Original. The chosen prompt is copied to the clipboard and VS Code Chat is opened when the editor exposes that command.

For Copilot, Codex CLI, Claude Code, or other assistant integrations, the reusable core in src/core can be wrapped by each platform's supported command, pre-submit hook, chat participant, or CLI workflow.

Recommended Assistant Type

For Senior Principal Prompt Architect work, use the strongest available reasoning-capable assistant in your target tool when the prompt needs repository awareness, implementation planning, review, security, testing, deployment, or architecture tradeoffs.

Use a lighter or faster assistant for low-risk prompt cleanup, copy editing, formatting, or shortening. The extension recommends an assistant type instead of naming a specific vendor model so suggestions do not become stale.

Commands

• Prompt Architect: Open Pre-Send Chat
• Prompt Architect: Optimize Prompt
• Prompt Architect: Review Prompt Quality
• Prompt Architect: Open Dashboard

Production Workflow

1. Start the extension with F5.
2. Use the Prompt Architect Activity Bar icon, the status bar item, Ctrl+Alt+P, or the command palette.
3. Open Prompt Architect: Open Pre-Send Chat.
4. Enter the chat you planned to send.
5. Select Use Optimized Prompt, Edit Optimized Prompt, or Ignore Suggestion and Use Original.
6. Paste the copied prompt into the target assistant chat.

The extension also registers an Activity Bar view and a status bar shortcut so it is visible after startup, not only after a command has been searched manually.

Development

Run tests:

npm.cmd run test

Run syntax checks:

npm.cmd run lint

Install locally in VS Code by opening this folder and pressing F5 from an extension host-capable VS Code environment.