Prompt Enhancer

Turn a rough thought into a prompt worth sending.

You know the one you were about to type:

I want you to fix the landing page, make it so so good

Prompt Enhancer rewrites that into a prompt that states a role, the task, the relevant context, the constraints, and the output format you want back- then puts it on your clipboard so you can paste it into whichever AI you actually use.

It invents nothing. If you didn't say which framework, it won't pick one; it lists what's unspecified so you can fill it in.

Works with any chat

Claude Code, ChatGPT, Copilot, Gemini, anything. The result goes to your clipboard and you paste it.

That's deliberate, and it's a platform limit worth being straight about: no VS Code extension can read or edit the text inside another extension's chat box. Those panels are private to the extension that owns them. So "select what you typed in the chat and press a key" isn't something this- or any- extension can do. The clipboard is the one bridge that works everywhere.

Bring your own key

No account, no sign-in, no server. You supply an API key for Anthropic, OpenAI, or Google AI, and the extension talks to that provider directly.

1. Prompt Enhancer: Set API Key from the command palette.
2. Paste your key. The provider is worked out from the key- and if the format is unfamiliar, you're asked rather than refused.
3. Pick a model from the list your key can actually use.

Keys live in your operating system's credential store, via VS Code's SecretStorage. One key per provider, so you can keep several and switch without re-pasting.

Three ways in

The panel

Click the Prompt Enhancer icon in the activity bar. Type or paste your rough prompt, pick a mode, hit Enhance. The result streams in as it's written, stays editable, and is copied to your clipboard when it finishes. Ctrl+Enter in the input runs it.

Use this when you want to see the result, tweak it, or try a different mode.

The shortcut

Ctrl+Alt+E (Cmd+Alt+E on macOS) anywhere- including with a chat panel focused. Type the rough prompt, press Enter, and the enhanced version is on your clipboard and open in a tab.

Use this when you're mid-flow and just want the text.

Rewriting a prompt that lives in a file

Select prompt text in a file and run Prompt Enhancer: Enhance Selection from the context menu or the palette. It's replaced in place, as a single undo step.

This is for files that contain prompts- CLAUDE.md, .github/copilot-instructions.md, a system prompt in your source, a scratch note. It is not for enhancing code: pointing it at a .tsx file will faithfully rewrite your component as a prompt, which is not useful to anyone.

If the request fails, you cancel it, or you edit the document while it's running, your buffer is left exactly as it was and the result opens in a separate tab instead.

Modes

Privacy

The text you send is sent to whichever provider your API key belongs to- Anthropic, OpenAI, or Google AI- and nowhere else. Which one follows entirely from the key you supply; set only an Anthropic key and your text only ever goes to Anthropic.

• The extension has no server. Nothing is proxied through the author.
• It collects no telemetry of any kind.
• The author never sees your text or your key.
• Your key is stored in the OS credential store via VS Code SecretStorage. It is never written to your settings, your workspace, or a log- the log output strips key shapes unconditionally.
• On OpenAI, requests are sent with store: false, so your text isn't retained in your OpenAI account after the response.

The only network requests made are to your provider: one to list the models your key can use, and one per enhancement.

Settings

No model IDs are built into this extension. They're discovered from your provider at runtime, so a newly released model works without waiting for an update.

Limits

• A prompt must be between 1 and 20,000 characters. That's what stops a whole-file paste becoming a surprise charge on your key.
• Requests time out after 30 seconds and can be cancelled at any point.
• The prompt template is tuned primarily against Anthropic. The other two providers use the same template and are supported, but their output isn't guaranteed to be identical in quality.

Known gaps

Being straight about what has and hasn't been exercised against a live API:

• The Anthropic path is the primary one and the template is tuned for it.
• The OpenAI and Google AI adapters are complete and unit-tested against their SDKs, but have not yet been run against the live APIs. If you hit something, please open an issue- it's more likely to be an untested assumption than a deep problem.

Requirements

VS Code 1.90 or later, and an API key from one of the three supported providers.