The Claude Code tips pack is a shipped cue config that teaches Claude Code inside Claude Code: 37 tip groups covering 128 words and phrases, from /compact to ultrathink to --print. Type one of its words in the prompt box, the word dims, and navigating to it shows the tip in the status line. No LLM call is involved; it's a pure static dictionary, instant and free.
What it teaches
The pack covers Claude Code's own vocabulary across slash commands, keybindings, flags, and features: plan mode, hooks, MCP, skills, CLAUDE.md, worktrees, model selection, permissions, session control, and more. Real entries, verbatim:
| You type | The tip says | Alternatives offered |
|---|---|---|
/compact | Summarize history when 'context limit' warning appears | /clear, /rewind |
/rewind | Undo everything - rolls back context AND file changes | /compact, /clear, Esc x2 |
ultrathink | Add ultrathink to prompt for max reasoning | deep thinking, think harder |
Shift+Tab | Double Shift+Tab enters plan mode for careful exploration | plan, careful |
opus | Use Opus for complex architecture - best reasoning, higher cost | sonnet, haiku, /model |
--print | Use --print flag for CI/CD non-interactive execution | ci, cd, pipeline |
Why teach a tool inside the tool
This pack is the working demonstration of continuous onboarding: agentic tools ship features faster than documentation can describe them, so the teaching surface moves into the prompt itself. You learn ultrathink at the moment you type it, while using it, and when Claude Code's behaviour changes, updating one config file updates every user's tips on their next keystroke. There is no separate documentation to go stale.
How it's scoped and gated
The pack declares on-host: [claude-code], so its trigger words never collide with vocabulary on other hosts; sibling packs exist for OpenCode, Gemini CLI, and the shell app. Display is gated by the tips-mode scalar, on by default: once you know every entry, turning tips off is the intended lifecycle, and flipping them back on after a feature ships is one line. To silence a single word instead, add it to the ignore: list.
Make your own pack
It's one CUE.md file of JSON tip groups, so the same shape works for your team's internal CLI, your project's jargon, or your own memory joggers: copy the folder, swap the words and tips, and it hot-reloads into the host it targets.
Related: What is tips mode? · Continuous onboarding · OpenCues in Claude Code · The claude-status blank · GitHub repo