Pause — Think & Refine Prompts

Stop firing off vague prompts. Pause adds a moment of intentional friction: it scores your prompt, asks one adaptive clarifying question at a time, and rewrites it using the target model's own prompting conventions.

build a todo app → score 2/100 → two questions → a rewritten prompt scoring 69/100

What it does

• Live prompt scoring — instant, local, zero latency.
• Adaptive clarification — one question at a time, each informed by your previous answers. Not a single-shot rewrite: it stops early when your prompt is already strong.
• Per-model formatting — the same intent, rewritten the way ChatGPT, Claude, Gemini, Grok, Perplexity, DeepSeek, or Mistral each respond best.

Usage

Open the ⏸ Pause panel from the activity bar (or the status bar item, or Ctrl+Alt+P) and type a rough prompt. The panel is a chat: it scores your prompt as you type, asks its clarifying questions as chat turns with clickable suggestions, and returns the rewrite with Send to Chat, Copy, and Insert at cursor buttons.

Every rewrite is copied to your clipboard automatically, so pasting into any AI panel — including ones VS Code sandboxes, like Claude Code or Codex — is a single Ctrl+V.

Settings

• pause.verbosity — concise (up to 3 questions, default) or thorough (up to 5).
• pause.targetModel — which AI's conventions to format the rewritten prompt for.
• pause.proxyUrl — endpoint AI calls route through.

Privacy

No API keys are stored in the extension, and your code is never uploaded. Only the prompt text and your answers to the clarifying questions are sent, and only to the proxy that performs the rewrite. See PRIVACY.md.

Requirements

The @pause chat participant requires an editor with the VS Code Chat API and a chat provider (e.g. GitHub Copilot Chat). The Refine Prompt commands work without it.

Pause also ships as a Chrome extension that adds prompt refinement to the AI sites you already use — and hands your conversation off to another AI when you hit a rate limit. pause-murex.vercel.app