Prompt Compiler

Select a prompt. Run Prompt Compiler: Optimize & Copy. Paste the result into your AI coding assistant.

Prompt Compiler turns a rough developer prompt into a clear, actionable one — then puts it straight on your clipboard.

Raw Developer Prompt → Prompt Compiler → Optimized Coding Prompt → Clipboard

It does not carry out the task in your prompt. It only rewrites the prompt.

Features

• One step. Select text, run the command, the result is already in your clipboard.
• Sidebar panel. Write and optimize a prompt right in the panel — no file needed.
• Keeps your language. A Vietnamese prompt produces Vietnamese output; English produces English; a mixed prompt keeps its natural mix. Technical terms (API, middleware, refactor, migration, …) are never translated.
• Preserves intent. No invented requirements, files, APIs, frameworks, or bug causes. When information is missing, the rewritten prompt asks the assistant to inspect the repository or check back with you.
• No prompt bloat. A short prompt stays short. It never bolts on "You are a senior software engineer…" or "Think deeply…".
• Your key, your provider. No account, no backend, no telemetry.
• Any OpenAI-compatible API. OpenAI, OpenRouter, LM Studio, Ollama, and compatible gateways.
• Model list loaded for you from the /models endpoint, picked through a filterable combo box.
• Automatic retry on transient provider failures (429/500/502/503/504) with 1s then 2s backoff.

The extension UI and its messages are currently in Vietnamese. Output language always follows your prompt.

Installation

From a .vsix file:

code --install-extension promptc-0.1.0.vsix

Or in VS Code: Extensions → … menu → Install from VSIX…

Build from source

npm install
npm run compile

Press F5 to launch the Extension Development Host.

Configuration

Open the Prompt Compiler icon in the Activity Bar → Cấu hình (Settings) tab.

One Lưu (Save) button persists everything, including the API key.

Test model sends one tiny request — any content coming back means the model works.

These fields also live in VS Code Settings under the promptOptimizer.* namespace. The API key does not — it stays in SecretStorage only.

Supported Providers

Anything that serves POST {baseUrl}/chat/completions in OpenAI format.

These four are one-click preset chips. For any other gateway, paste its URL.

Usage

From the editor

1. Write a prompt in any file.
2. Select it.
3. Ctrl+Shift+P → Prompt Compiler: Optimize & Copy.
4. Ctrl+V into your AI coding assistant.

From the sidebar

1. Open the Prompt Compiler icon in the Activity Bar.
2. Type into the prompt box, or click Từ selection to pull in the current editor selection.
3. Click Tối ưu & Copy, or press Ctrl+Enter.
4. The result appears and is already on the clipboard.

Commands

Example

Input:

đổi tên biến x thành userId

Output:

Đổi tên biến `x` thành `userId`.

- Cập nhật tất cả reference liên quan.
- Không thay đổi behavior.
- Kiểm tra các lỗi compile/type-check sau khi thay đổi.

A short prompt stays short. It is never inflated into 500 words.

Privacy

Prompts go directly from the VS Code extension to the LLM provider you configured. The extension runs no backend of its own and collects no prompts, source code, or API keys.

If your third-party provider has its own policy, checking it is up to you.

Precisely what the extension does:

• Sends: only the fixed meta prompt and exactly the text you selected (or typed into the panel). No other source code, no environment variables, no git history.
• Stores: the API key in VS Code SecretStorage; settings in VS Code settings. Nothing else.
• Logs: the Output channel records metadata only — URL, model name, HTTP status, body size, API key length. Never prompt content, response content, or the key itself.
• Telemetry: none.

Troubleshooting

Run Prompt Compiler: Show Log from the Command Palette to see request details.

Development

npm install
npm test                 # typecheck + lint + unit tests
npm run lint             # eslint with type-aware rules
npm run typecheck        # tsc --noEmit
npm run test:unit        # unit tests (node:test)
npm run test:coverage    # unit tests with coverage
npm run test:integration # runs inside real VS Code against a mock LLM server
npm run bundle           # esbuild → dist/extension.js
npm run package          # build the .vsix

The shipped extension is a single esbuild bundle (dist/extension.js), so the VSIX stays around 24 KB and activation stays fast. npm run watch rebuilds it on change; press F5 to launch the Extension Development Host.

Releasing

# 1. update CHANGELOG.md, then:
npm version patch    # or minor / major — writes package.json and tags
git push --follow-tags

The Release workflow verifies the tag matches package.json, runs the full test suite, packages the VSIX, and attaches it to a GitHub release. No secrets required.

Publishing to the Marketplace stays manual: download the VSIX from the release and upload it at Manage Publishers & Extensions.

Architecture:

src/
├── extension.ts              # activate, register commands and the view
├── commands/                 # VS Code layer: selection, notifications, clipboard
├── core/                     # pure logic, NEVER imports vscode
│   ├── optimizer.ts          # pipeline orchestration
│   ├── promptBuilder.ts      # meta prompt + language hint
│   ├── responseNormalizer.ts # strips an outer code fence
│   ├── languageDetector.ts   # vi/en/mixed/unknown heuristic
│   └── errorMessages.ts      # error code → user-facing message
├── providers/                # HTTP layer, NEVER imports vscode
├── config/                   # settings resolver + SecretStorage
├── constants/metaPrompt.ts   # the meta prompt (verbatim from SPEC-001)
└── ui/configView.ts          # sidebar webview

core/ and providers/ do not depend on the VS Code API, so unit tests run directly under node --test without booting VS Code.

License

MIT