pqc-beta

pqc-beta is a VS Code extension (in early beta) that helps people with little or no prompt-engineering experience understand whether a prompt is likely to produce useful AI responses.

The extension runs local heuristic analysis by default, and seamlessly integrates with GitHub Copilot to run deep, cost-optimized LLM audits out-of-the-box with zero configuration required.

What it measures

The first version scores seven practical dimensions:

• Clarity: whether the prompt has a clear task and readable instructions.
• Specificity: whether it includes concrete details, limits, and success criteria.
• Context: whether it explains audience, goal, background, role, or scenario.
• Constraints: whether it states must-have requirements, exclusions, tone, and trade-offs.
• Output format: whether it requests Markdown, JSON, tables, schemas, bullets, or another structure.
• Examples: whether it includes sample input/output or few-shot examples.
• Safety and boundaries: whether it handles privacy, citations, uncertainty, or domain-sensitive guidance.

Features

• Built-in Copilot Integration: Enhances your main Copilot chat thread with automated prompt auditing via the #pqc-check tool. No extra setup required.
• Dedicated Chat Participant: Use @pqc-beta for a dedicated, sticky auditing workspace.
• One-Click Audit Bar: Look for the Sparkle icon ✨ in the Copilot Chat title bar to jump straight into an audit.
• Cost-Optimized LLM Routing (0X Quota): Automatically detects Copilot background models (e.g., gpt-4o-mini, gemini-1.5-flash) and routes prompt audits to the cheapest 0X cost model so you never burn your premium high-end LLM quota.
• Antigravity & Chat Forwarding: Configure a dynamic destination (pqcBeta.forwardTarget) to send finalized, audited prompts directly to Antigravity or GitHub Copilot with a single click.
• Smart Auditing Dialog: Displays active model information and costs, featuring an "Always Allow" option for frictionless interaction.
• Keyboard Shortcut: Press Cmd+Alt+P (macOS) or Ctrl+Alt+P (Windows/Linux) to audit your current editor content instantly.
• Analyze selected text or the whole active editor.
• See an overall 0-100 score, grade, word count, sentence count, and readability estimate.
• Review metric-by-metric feedback with positives and suggested improvements.
• Insert a local improved prompt draft directly into the editor.
• Store custom OpenAI keys in VS Code SecretStorage as an optional advanced fallback.
• Insert beginner-friendly prompt templates for coding, product, research, data, writing, API documentation, testing, security, and performance tasks.
• Export the latest prompt quality report as Markdown.
• See the latest prompt rating as a small VS Code status-bar icon.
• Review the last 10 prompt ratings before moving to the next prompt or chat.
• Use the dedicated pqc-beta activity bar view for results.

Roadmap

The next major improvement is a local-first contextual analyzer that can resolve follow-up prompts such as fix this or what about memory leaks? using chat, editor, and prompt history context. See Contextual Analyzer Plan.

Recommended API model

The recommended product model leverages your existing GitHub Copilot connection directly:

1. Zero Configuration: No API keys, registration, or accounts are required. The extension runs out-of-the-box using local rules and your active Copilot session.
2. 0X Cost Model Routing: Prompt auditing is routed to highly efficient, cost-free background models (like gpt-4o-mini or gemini-1.5-flash), protecting your premium high-end LLM quota.
3. Optional Custom Keys: Developers who wish to audit prompts using custom setups can optionally set their own OpenAI API key.

This keeps the first marketplace release simpler, cheaper, and completely privacy-transparent.

How to run locally

1. Install dependencies:

   npm install
   

2. Compile the extension:

   npm run compile
   

3. Press F5 in VS Code to open an Extension Development Host.

4. Open or create a file containing a prompt, select the prompt text, and run pqc-beta: Analyze Prompt from the Command Palette or editor context menu.

5. Click the Prompt score in the status bar, or run pqc-beta: Show Recent Prompt Ratings, to review recent prompt ratings before starting the next chat or prompt.

6. Optional: run pqc-beta: Set OpenAI API Key, then pqc-beta: AI Review Prompt for deeper model-based feedback.

How to package for publishing

1. Update the publisher field in package.json to your VS Code Marketplace publisher ID.

2. Install VSCE if it is not already available:

   npm install
   

3. Build a .vsix package:

   npm run package
   

4. Publish through the Visual Studio Marketplace publisher portal or run vsce publish after logging in.

Configuration

• pqcBeta.minimumGoodScore: controls the score threshold used in the status notification.
• pqcBeta.includeImprovedDraft: toggles the improved draft section in the metrics view.
• pqcBeta.showStatusBarRating: toggles the small status-bar rating icon.
• pqcBeta.maxRecentPrompts: controls how many recent prompt ratings are stored locally.
• pqcBeta.forwardTarget: choice of target AI assistant for audited prompt forwarding (copilot or antigravity).
• pqcBeta.alwaysAllowLlmAudits: bypasses the cost-confirmation dialog for Copilot LLM audits once allowed.
• pqcBeta.ai.enabled: enables opt-in AI review commands (OpenAI).
• pqcBeta.ai.model: controls the OpenAI model used for AI reviews.
• pqcBeta.ai.requireConfirmationBeforeSending: asks before sending prompt text externally.
• pqcBeta.ai.cacheReviews: reuses AI reviews for identical prompt text.

Limitations

This is a heuristic quality checker plus optional model review, not a guarantee of downstream model performance. Treat scores as coaching signals and validate important prompts with real outputs, user feedback, and domain review.

The extension can show ratings for prompts analyzed inside VS Code. It cannot automatically detect when an external chat website is closed unless that chat text is represented in the editor and the user runs an analysis command.