Lens — Prompt Improver

Turn rough AI prompts into specific, structured, constrained ones. One keystroke.

Lens is a VS Code extension that sits between you and your AI coding assistant (Copilot, Cursor, Codex, Claude Code, ChatGPT). Select a rough prompt, press ⌘K L, and Lens rewrites it in place — adding specificity, structure, and safety guards so the downstream model gives you a better answer with fewer retries.

Opt-in per action. Never intercepts silently. Works with any OpenAI-compatible API.

Why

Bad prompts cause expensive retries. You ask vaguely, get a mediocre answer, refine, ask again. That's three LLM calls where one would have sufficed. Lens fixes the prompt before it hits the expensive model.

Before:

can you look into pricing and tell me what files are important

After (⌘K L):

In the current repository, identify the files most relevant to pricing logic. Rank them by importance based on: (1) file size and complexity, (2) how many other files reference them, (3) recency of changes. Return a ranked list with a one-sentence justification per file. Do not modify any files.

Same intent. Vastly better answer downstream. Roughly 15–30% cheaper in downstream LLM spend once you factor in avoided retries.

Install

Install from the VS Code Marketplace or:

code --install-extension your-publisher.prompt-lens

On first use, Lens will prompt you to paste an API key. Any OpenAI-compatible key works. The key is stored in VS Code's SecretStorage — never in settings, never synced to git.

Usage

Improve a prompt in place — ⌘K L (Mac) / Ctrl+K Ctrl+L (Win/Linux)

1. Select the rough prompt text (in any file — .md, code, comments, notebook cells)
2. Press ⌘K L
3. Lens replaces it with the improved version
4. ⌘Z to undo if you don't like it

No selection? Lens grabs the current line.

Compose a new prompt — ⌘K ⇧L (Mac) / Ctrl+K Ctrl+Shift+L (Win/Linux)

For sending prompts to chat panels (Copilot Chat, etc.) where in-place editing isn't feasible:

1. Press ⌘K ⇧L from anywhere
2. Type or paste your rough prompt
3. Review the improvement
4. Hit Enter — Lens copies to clipboard (default) or opens Copilot Chat with the improved prompt

Configuration

Open Settings → Extensions → Lens.

Commands

• Lens: Improve Selected Prompt — the main rewrite action
• Lens: Compose New Prompt — panel for chat-panel workflows
• Lens: Set API Key — securely store your API key
• Lens: Clear Stored API Key — remove the stored key
• Lens: Show Session Stats — how many rewrites, acceptance rate, tokens

Cost

Lens uses a small model for rewrites (default gpt-4o-mini). Each rewrite is roughly $0.0001 — one hundredth of a cent. The downstream model call is where the real money is, and that's where you save 15–30% by having a better prompt.

Privacy

• Prompts you rewrite are sent to the API endpoint you configure (default: OpenAI). Lens is a client — it does not proxy anything through any Lens-owned server.
• File contents are never sent. Lens only sends the prompt text you selected, plus optional metadata (current file language, workspace-relative filename) if lens.includeEditorContext is on.
• API key is stored in VS Code's SecretStorage (OS keychain on Mac/Linux, Windows Credential Manager on Windows).
• Telemetry is opt-in and off by default. When enabled, only anonymous counts (rewrite attempted, accepted, rejected) are sent — never prompt contents.

What Lens is NOT

• Not automatic. It only fires when you press the hotkey. Predictable > clever.
• Not a code assistant. It doesn't answer your prompt — it just improves it before you send it somewhere that will.
• Not proprietary to any provider. Any OpenAI-compatible endpoint works.

Building from source

git clone <this-repo>
cd prompt-lens
npm install
npm run compile

To try it locally: open the folder in VS Code, press F5. This launches an Extension Development Host with Lens loaded.

To package a .vsix:

npm run vsce:package
# → prompt-lens-0.1.0.vsix
code --install-extension prompt-lens-0.1.0.vsix

Testing

npm test

License

MIT. See LICENSE.