Prompt Queue + Usage Monitor

Queue prompts for later · Monitor AI token usage · Browse Claude slash commands

A VS Code extension built for developers who use Claude Code, Copilot, or any AI assistant with rate limits. When you're blocked, queue your next prompt — the extension delivers it automatically when the time comes.

What it does

Installation

From VSIX (recommended while in preview)

code --install-extension prompt-queue-usage-monitor-0.3.0.vsix

From source

git clone <repo>
cd send_later_extension
npm install
npm run compile

Press F5 to open the Extension Development Host.

Usage

Queue a prompt when you're rate-limited

The fastest path: hit a rate limit → run the smart command → done.

1. Copy the rate-limit message to your clipboard (e.g. "Try again in 1h 30m" or "Resets at 22:00").
2. Open the Command Palette (Cmd+Shift+P / Ctrl+Shift+P).
3. Run PromptQueue: I'm Rate Limited — Queue for Later.
4. The extension parses the reset time automatically and pre-fills the delay. Confirm and queue.
5. When the timer fires, the prompt is sent directly to your Claude terminal — or written as a .md file in .prompt-queue/ if no terminal is open.

You can also queue prompts manually:

• Queue Prompt (Send Later) — type or paste your prompt, set a delay in minutes (30) or a target time (22:00)
• Queue From Clipboard — queues the current clipboard content
• Queue From Current Editor — queues the whole file or your current selection

The queue survives VS Code restarts (persisted in global state).

Monitor token usage

Open the Prompt Queue Activity Bar icon → click the Usage tab.

The panel shows:

• A progress bar for your last 5 hours of token usage
• A progress bar for your last 7 days
• Provider status (which data source is active)

No setup required: the Claude Local provider reads ~/.claude/projects/*.jsonl automatically. For OpenAI or Anthropic admin data, add your keys (see below).

Calibrate limits from claude.ai

1. Open claude.ai and note the percentage shown (e.g. "68% used").
2. Run Usage: Calibrate Limits from claude.ai %.
3. Enter the percentage — the extension back-calculates your token ceiling and stores it.

Set API keys (optional)

Keys are stored in VS Code SecretStorage (OS keychain) — never written to settings.json.

OpenAI — requires an organization admin key (sk-org-…), not a project key:

Usage: Set OpenAI API Key (Secret)

Anthropic — requires an admin API key from console.anthropic.com:

Usage: Set Anthropic Admin API Key (Secret)

Browse Claude slash commands

Open the Commands tab in the Activity Bar panel.

• All .claude/commands/**/*.md files in your workspace are scanned automatically
• Commands are grouped by folder (category)
• Click a command card → slash path copied to clipboard (e.g. /workflows/code-review)
• Search filters by name and description with live highlight
• Open button opens the full command file in the editor
• ↻ button rescans the workspace

All commands

Configuration

How it works

Prompt delivery

Every 60 seconds (and on VS Code focus) the processor checks all queued items. For each due item:

1. Finds your Claude terminal — searches by name hint → "Claude" → any terminal containing "claude".
2. If found → sends the prompt via terminal.sendText() (bracketed paste, ESC-stripped).
3. If not found → creates a new terminal, saves prompt to a temp file, runs claude "$(cat file)", cleans up.
4. Writes a .md file to <outputDir>/ via workspace.fs (works on Remote / WSL / SSH).

If VS Code is closed when a prompt is due, it is delivered on the next activation.

Rate-limit parser

The parser recognises patterns from Claude Code, Copilot, and generic API messages:

A 5-minute safety buffer is added automatically to every parsed delay.

Usage providers

Security

• No secrets in settings.json — API keys use vscode.SecretStorage (OS keychain on macOS, Credential Manager on Windows, libsecret on Linux).
• No telemetry — nothing is collected or sent.
• No hard-coded credentials.
• Network requests go only to api.openai.com and api.anthropic.com, and only when keys are configured.

Development

npm install          # install deps
npm run watch        # compile in watch mode
# Press F5 → Extension Development Host
npm test             # run unit tests (pure, no VS Code instance needed)
npx vsce package     # build .vsix

Project structure

src/
  extension.ts                         ← Activation, command wiring
  queue/
    QueueStore.ts                      ← Persistent queue (globalState Memento)
    QueueProcessor.ts                  ← 60s interval + terminal delivery
  usage/
    IUsageProvider.ts                  ← Interface + shared types
    ClaudeLocalProvider.ts             ← ~/.claude/projects/*.jsonl reader
    OpenAIUsageProvider.ts             ← Admin usage API
    AnthropicUsageProvider.ts          ← Admin usage API
    LocalEstimateProvider.ts           ← chars/4 fallback
    UsageService.ts                    ← Aggregation + caching
  ui/
    QueueWebviewProvider.ts            ← Prompt Queue panel
    UsageWebviewProvider.ts            ← Usage Monitor panel
    ClaudeCommandsWebviewProvider.ts   ← Commands browser
  util/
    time.ts                            ← Time helpers + rate-limit parser
    crypto.ts                          ← 8-char hex ID generator
    fs.ts                              ← workspace.fs helpers
  test/
    suite/
      time.test.ts                     ← Pure unit tests
      queue.test.ts                    ← MockMemento queue tests

License

MIT