PromptRev

Enhance your AI prompts before you send them — directly in VS Code Chat.

Type @rev your rough prompt and get a polished, context-aware version back with a diff showing exactly what changed. No API key required. No background process. Works with whatever model you already have.

Getting started

Install the extension
Open VS Code Chat (Ctrl+Alt+I / Cmd+Alt+I)
Type @rev followed by your prompt

@rev fix the bug in my auth module

PromptRev enhances the prompt, shows you what changed, and gives you three options: Send Improved, Edit, or Send Original.

Modifiers

Append a /modifier to control how your prompt is rewritten:

Command	What it does
`@rev`	Clarity, structure, and missing context — everyday default
`@rev /rb`	Rule-based enhancement — no model required, auto-accept
`@rev /deep`	Chain-of-thought framing, edge cases, structured output
`@rev /architect`	System design framing — trade-offs, components, scalability
`@rev /critic`	Adversarial review — failure modes, security, assumptions
`@rev /spell`	Grammar and spelling only, no restructuring
`@rev /spec`	Converts a vague idea into a structured spec

Prompt templates

Browse and fill reusable prompt structures with @rev /template or Cmd+Shift+T:

Template	Best for
`/bug-fix`	Root cause + fix + regression test
`/code-review`	Structured review with severity ratings
`/architecture`	System design with constraints and trade-offs
`/refactor`	Targeted refactor with goals and risk notes
`/test-generation`	Tests with coverage goals and edge cases
`/explain-code`	Code explanation at a specified audience level
`/pr-description`	PR title, summary, and test plan
`/incident-postmortem`	Timeline, root cause, and action items

Each template walks you through filling in variables step-by-step. If you have text selected in the editor, PromptRev will offer to use it as context automatically.

Keybinding — enhance selected text

Select any text and press Cmd+Shift+E (Mac) / Ctrl+Shift+E (Windows/Linux) to enhance it without opening chat. Works in .md files, scratch pads, code comments — anywhere you draft prompts.

Configuration

{
  "promptrev.domainContext": "React + TypeScript, Node.js, PostgreSQL",
  "promptrev.maxHistory": 100,
  "promptrev.modifiers": {
    "backend": {
      "systemPrompt": "Rewrite for a Node.js + Express + Prisma backend. Include TypeScript, JWT auth, error handling, and input validation.",
      "acceptMode": "diff"
    }
  },
  "promptrev.templates": {
    "my-api-endpoint": {
      "name": "API Endpoint Design",
      "template": "Design a {{method}} endpoint at {{path}} for {{purpose}}.",
      "variables": [
        { "name": "method", "placeholder": "POST", "required": true },
        { "name": "path", "placeholder": "/api/users", "required": true },
        { "name": "purpose", "placeholder": "creating a user", "required": true }
      ]
    }
  }
}

Setting	Default	Description
`promptrev.domainContext`	`""`	Tech stack injected into every enhancement
`promptrev.maxHistory`	`100`	Max history entries kept (1–500)
`promptrev.modifiers`	`{}`	Custom or overridden modifier definitions
`promptrev.templates`	`{}`	Custom template definitions

Team config

Commit a .promptrev.json at the repo root to share modifiers and templates across your team:

{
  "domainContext": "React + TypeScript, Node.js, PostgreSQL",
  "modifiers": {
    "backend": {
      "systemPrompt": "Rewrite for our Node.js + Prisma stack.",
      "acceptMode": "diff"
    }
  },
  "templates": {
    "incident-report": {
      "name": "Incident Report",
      "template": "Write a postmortem for {{service}} — duration: {{duration}}, impact: {{impact}}.",
      "variables": [
        { "name": "service", "placeholder": "payments-api", "required": true },
        { "name": "duration", "placeholder": "47 minutes", "required": true },
        { "name": "impact", "placeholder": "checkout unavailable for EU users", "required": true }
      ]
    }
  }
}

Requirements

VS Code 1.85.0 or later
GitHub Copilot or another VS Code language model extension (recommended)

If no model is available, :rb mode still works via rule-based enhancement — no model required.