PromptCheckup

A multilingual prompt diagnosis and optimization workflow built with Dify Chatflow.

PromptCheckup is a Chinese-first, multilingual prompt diagnosis and optimization tool built with Dify Chatflow and a local Web UI.

PromptCheckup 是一个基于 Dify Chatflow 的多语言 Prompt 体检工具，支持结构预检、意图分流、任务分类、 结构化诊断、风险评分、Prompt 优化、多轮追问调整和修改表单后的重新诊断。

中文名：Prompt 体检医生

Project Overview

PromptCheckup v0.3.0 includes four parts:

• Dify Chatflow template: dify/prompt-checkup.yml
• Local React Web UI
• Local Node Dify API wrapper
• Vercel API Functions for hosted demo deployment

The Web UI calls /api/chat. In local development this is served by the local Node wrapper; on Vercel it is served by Vercel API Functions. Both paths call your own Dify App API server-side. PromptCheckup does not include model credits, hosted inference, a database, login, or cloud history.

PromptCheckup v0.3.0 adds Vercel demo deployment support while keeping the local Web UI workflow available. The hosted demo path uses Vercel API Functions, demo access-code protection, and input length limits. See docs/vercel-deployment.md. The initial Vercel demo uses the Dify Chat API in blocking mode, so complex reports can take a long time and may hit upstream or serverless timeout limits.

Features

• Importable Dify Chatflow template
• Local React Web UI
• Local Node Dify API wrapper
• Vercel API Functions for hosted demos
• Real prompt diagnosis through the user's own Dify App API
• Multi-turn follow-up adjustment
• Re-diagnose current form
• New session
• Local draft persistence
• Chinese / English / Japanese UI
• Copy full report
• Copy last answer
• Download Markdown
• API key kept on local backend, not in browser
• Demo access code support for hosted Vercel sharing

Architecture

The Dify Chatflow core follows this path:

User Input
-> Interaction Intent Routing
-> Precheck
-> Validity Gate
-> Task Classification
-> Non-Prompt Reply / Brief Diagnosis / Structured Diagnosis
-> Score Calculation
-> Final Report
-> Follow-up Adjustment

The local v0.2 runtime adds:

Browser Web UI
-> Local /api/chat
-> Local Node wrapper
-> Dify App API /chat-messages

Detailed notes are in docs/architecture.md, docs/web-ui.md, and docs/local-wrapper.md.

Screenshots

Quick Start

Local Development

1. Clone this repository.
2. Import dify/prompt-checkup.yml into Dify.
3. Configure the model provider inside your own Dify workspace.
4. Publish the Dify app.
5. Create a Dify App API Key.
6. Copy .env.example to .env.
7. Fill the local environment variables:

DIFY_API_BASE_URL=https://api.dify.ai/v1
DIFY_API_KEY=your_dify_app_api_key_here
DIFY_USER=local-user
SERVER_PORT=8787
DEMO_ACCESS_CODE=change_me_for_public_demo

8. Install dependencies and start the local app:

npm install
npm run dev

9. Open:

http://localhost:5173

You must use your own Dify App API Key. Model usage and cost are handled by your own Dify workspace and model provider configuration. Do not commit .env.

Vercel Demo Deployment

1. Import the GitHub repository into Vercel.
2. Use the repository root directory ./.
3. Confirm build settings:

Framework Preset: Vite
Build Command: npm run build --workspace web
Output Directory: web/dist

4. Configure Vercel Environment Variables:

DIFY_API_BASE_URL=https://api.dify.ai/v1
DIFY_API_KEY=your_dify_app_api_key_here
DIFY_USER=local-user
DEMO_ACCESS_CODE=change_me_for_public_demo

5. Deploy and open the generated Vercel URL.
6. Enter the configured demo access code in the Web UI before running diagnosis.

Store real values only in Vercel Environment Variables. Do not commit .env, server/.env, Dify keys, model provider keys, or real demo access codes to GitHub.

Import into Dify

In the Dify console, import:

dify/prompt-checkup.yml

After import, check the model provider configuration and publish the app before using the local Web UI. The exported Flow does not contain a public API key.

Usage

1. Fill the prompt diagnosis form.
2. Click Start Diagnosis.
3. Read the Markdown diagnosis report.
4. Use follow-up messages to adjust the output.
5. Use Re-diagnose current form after editing the form fields.
6. Use Copy full report, Copy last answer, or Download Markdown.
7. Start a New Session when you want to clear the conversation while keeping the form available.

Test Cases

Test notes are in docs/test-cases.md. Example inputs are stored in examples/.

Roadmap

See docs/roadmap.md. Vercel deployment notes are in docs/vercel-deployment.md.

Security

• DIFY_API_KEY is only read by the local backend or Vercel API Functions.
• The browser calls /api/chat.
• The frontend must not expose a Dify API Key.
• Hosted demo requests send the demo access code through x-demo-access-code.
• .env is ignored by git.
• Do not commit .env, Dify App API keys, model provider keys, or GitHub tokens.
• The Vercel demo access code is lightweight protection for preview sharing, not full authentication, per-user quota, or billing.

Known Limitations

• v0.2 does not provide dedicated optimized / advanced prompt copy buttons.
• Dify currently returns optimized / advanced prompts as natural-language Markdown sections, not stable machine-readable fields.
• Dedicated optimized / advanced prompt copy can return after a future structured output version adds stable fields or explicit Markdown markers.
• Review-depth modes may not always produce strongly differentiated results until the Dify Flow prompt logic is upgraded.
• v0.3 Vercel demos initially use Dify blocking mode. Complex reports may time out; retry with a shorter prompt, a lighter review mode, or try again later.
• Streaming responses and Dify Flow performance optimization are planned future improvements.

License

MIT License. See LICENSE.