PromptForge
Evaluate, version and improve your LLM prompts — without leaving VS Code.
PromptForge is a VS Code extension that brings a full prompt engineering workflow directly into your editor. Write a .prompt file, evaluate it with an AI judge, and get actionable improvement suggestions in seconds.
No server setup. No API keys required. Just install and go.
Requirements
PromptForge uses the language models already available in your VS Code. You need one of the following:
- GitHub Copilot (Free, Pro, or Pro+) — Install
- Claude for VS Code — Install
- Groq API key (free) — as a fallback if no model is available
If no model is detected, PromptForge will guide you through the setup automatically.
Features
AI-powered evaluation
Every .prompt file is evaluated across 5 dimensions:
- Coherence — does the prompt have a clear, actionable instruction?
- Precision — is it specific enough to produce consistent outputs?
- Tone — is the tone appropriate for the context?
- Safety — is the prompt free from harmful instructions?
- Completeness — does it include a task, context, and expected output format?
Surgical improvement suggestions
When a dimension scores below 7, PromptForge generates a targeted rewrite that fixes only what's broken — leaving what already works untouched. Every change is explained and linked to the dimension it addresses.
Version history
Every evaluation automatically saves a snapshot of your prompt. Browse the full version tree in the sidebar, and restore any previous state with one click.
Apply improvements in one click
Accept a suggested improvement directly into your editor, ready for your next evaluation cycle.
How to use
- Create a file with the
.prompt extension
- Write your prompt — use
{variable} for template placeholders
- Press
Ctrl+Alt+E or click the beaker icon in the editor title bar
- Review the diagnosis in the PromptForge panel
- Click Suggest improvement if any dimension scores below 7
- Click Apply in editor to accept the suggestion
Using Groq as fallback
If you don't have Copilot or Claude installed, you can use a free Groq API key:
- Get your free key at console.groq.com
- Open VS Code Settings (
Ctrl+,)
- Search for
promptforge.groqApiKey
- Paste your API key
Commands
| Command |
Description |
PromptForge: Run Evaluation |
Evaluate the active .prompt file |
PromptForge: Show Version |
Preview a version from the history |
PromptForge: Restore Version |
Restore a previous version to the editor |
Keyboard shortcuts
| Shortcut |
Action |
Ctrl+Alt+E |
Run evaluation on the active .prompt file |
Supported models
PromptForge automatically selects the best available model in this order:
- Claude Sonnet (via Copilot)
- GPT-4o (via Copilot)
- GPT-4.1 (via Copilot)
- Any other available model via
vscode.lm
- Groq llama-3.3-70b (fallback with API key)
The quality of the evaluation depends on the model — larger models produce more accurate and detailed diagnostics.
Tech stack
- Extension — TypeScript + VS Code API
- Language models —
vscode.lm (Copilot, Claude) or Groq API
- Storage — SQLite via sql.js
License
MIT