OpenCues patches Claude Code so its prompt box helps you write better prompts: rewrite and extend your instruction inline before you submit it, cycle word alternatives, and learn Claude Code features from tips that surface as you type them.
Install
Install the CLI with npm install -g opencues, then run opencues install claude-code (full details in the repo README). The installer creates a separate patched copy of Claude Code, so your normal claude command is never touched; launch the patched one with opencues run claude-code. You will need an LLM API key (Cerebras recommended; Groq, OpenAI, Anthropic, Gemini, OpenRouter, and local Ollama also work), or you can route through your existing Claude subscription.
What you get in the prompt box
- Improve the prompt before you press Enter.
improve prompt _rewrites your draft instruction.add more detail _expands it.add a paragraph about security _extends it in place. A sharp instruction first time saves the expensive agent round trips that follow a vague one. - Swap words without retyping. Words with alternatives dim as you compose;
Ctrl+Alt+←/→walks them andCtrl+Alt+↑/↓cycles a sharper word in place. - Learn features as you use them. Type a feature word like
ultrathinkand navigating to it shows what it does in the status line. No docs tab. - Keep the prompt clean. Start
agentically correct spelling _and every pause in typing tidies your draft, each fix dimmed and revertable. - Without leaving the session. System blanks work in the prompt box too:
volume 40 _sets your volume mid-session,weather _checks conditions.
Everything is configured by markdown files under ~/.cues/, hot-reloaded as you type. Self-healing installs rebuild the patched copy automatically whenever the runtime source changes.
Related: What is a cue? · Which models can I use? · The Claude Code tips pack · The claude-status blank · GitHub repo