Prompt Craft

Prompt Craft (VS Code Extension)

Prompt Craft generates concise, AI-ready prompts from your codebase. It converts diagnostics and project context into high-signal prompts designed to work well with Copilot, Cursor, and ChatGPT.

Install

• VS Code Marketplace: mohitmishra.prompt-craft
  • https://marketplace.visualstudio.com/items?itemName=mohitmishra.prompt-craft
• From file: install prompt-craft-0.0.1.vsix (Extensions → … → Install from VSIX)

What it does

• Error prompts: Transforms Problems (Diagnostics) or pasted stack traces into structured prompts with Context, Error Details, Code Snippet, Goal, and Constraints.
• Project prompts: Scans package.json, README.md, and up to ~100 code/text files to infer tech stack, dependencies, and architecture signals; emits domain-aware tasks.
• Multi-provider LLM support: Choose from multiple AI providers:
  • Groq (llama3-8b-8192, llama3-70b-8192, mixtral-8x7b-32768)
  • OpenAI (gpt-4, gpt-4-turbo, gpt-3.5-turbo)
  • Azure OpenAI (with API key, Azure CLI, or Managed Identity authentication)
• Provider management: Easy provider switching via status bar, health checks, and configuration UI

Commands

Core Commands:

• Prompt Craft: Analyze Project and Generate Prompts - Analyze your project and generate AI prompts
• Prompt Craft: Generate Error Prompt - Create prompts from errors or diagnostics

Provider Management:

• Prompt Craft: Select LLM Provider - Choose your AI provider (Groq, OpenAI, Azure OpenAI)
• Prompt Craft: Test Provider Connection - Verify provider connectivity and health
• Prompt Craft: Show Provider Status - View all configured providers and their status
• Prompt Craft: Reload Providers from Settings - Refresh provider configuration

Status Bar:

• Click the provider icon in the status bar for quick provider switching
• Visual indicators show provider health (active/warning/error)

Quick start (development host)

1. Open this folder in VS Code
2. npm install
3. npm run compile
4. F5 to start the Extension Development Host

In the Dev Host, open a workspace to analyze (e.g., demo/ecom-backend), then run the commands from the Command Palette.

Provider Setup

Groq (Free Tier Available)

1. Get API key from groq.com
2. Settings → promptCraft.groqApiKey
3. Or set environment variable: GROQ_API_KEY

OpenAI

1. Get API key from platform.openai.com
2. Settings → promptCraft.openai.apiKey
3. Settings → promptCraft.openai.model (default: gpt-4-turbo)

Azure OpenAI

Option 1: API Key Authentication

1. Settings → promptCraft.azureOpenAI.endpoint (e.g., https://your-resource.openai.azure.com)
2. Settings → promptCraft.azureOpenAI.apiKey
3. Settings → promptCraft.azureOpenAI.deploymentName
4. Settings → promptCraft.azureOpenAI.authMethod = apiKey

Option 2: Azure CLI Authentication (No API Key Needed)

1. Install Azure CLI
2. Run az login in terminal
3. Settings → promptCraft.azureOpenAI.authMethod = azureCli
4. Configure endpoint and deployment name

Option 3: Managed Identity (Role-Based, Azure VMs)

1. Assign Managed Identity to your Azure VM/App Service
2. Grant "Cognitive Services User" role
3. Settings → promptCraft.azureOpenAI.authMethod = managedIdentity
4. Configure endpoint and deployment name

The Output shows "Generation: LLM-enhanced" when a provider is used; otherwise "Generation: Template".

Demo workspace

• demo/ecom-backend: minimal Express + MongoDB project with an intentional error in src/app.js and a crash route at /boom.
• Try both flows:
  • Diagnostics-based error prompt: uncomment const broken = ; → Problems shows an error → run Generate Error Prompt.
  • Stack-based error prompt: run the app, hit /boom, copy the stack → run Generate Error Prompt and paste.

Architecture and heuristics

• Tech stack detection via package.json (TypeScript vs JavaScript)
• Architecture signals from folder layout (src/, controllers/, models/, routes/)
• Domain inference (e.g., e-commerce) from README and filenames
• Lightweight extraction of controller function exports and model/schema fields

Security and privacy

• No hard-coded keys. Keys are stored in user/workspace settings or read from the environment.
• No eval or arbitrary code execution.
• Reads only within the open workspace.
• When LLM is enabled, requests include a compact analysis JSON (tech stack, dependencies, README summary, architecture summary, limited controller/model summaries). Do not include sensitive content in README or code if this is a concern.

Troubleshooting

• Commands don’t appear: ensure you are in the Extension Development Host; open a workspace with a README.md or run the command explicitly.
• LLM not used: set promptCraft.groqApiKey in the Dev Host Settings, or launch VS Code from a shell with GROQ_API_KEY; check for “Generation: LLM-enhanced”.
• Snippet missing: paste a stack trace that includes a resolvable path and line.

Development

• Code: TypeScript in src/, compiled to out/
• Key files:
  • src/extension.ts: activation, commands, output
  • src/analyzer.ts: workspace scanning and heuristics
  • src/templates.ts: templates and helper builders
  • src/llm.ts: Groq wrapper and prompts
  • src/domain.ts: domain inference and simple code extraction

Distribution

• Marketplace: https://marketplace.visualstudio.com/items?itemName=mohitmishra.prompt-craft
• Local VSIX: npx vsce package → install the generated .vsix

License

See LICENSE.