31 Hardcore Claude Code Tricks from Anthropic’s Ado Kukic – Save Them All

Hello, I’m Zhijian‑Jun! In December, Anthropic’s developer‑relations lead Ado Kukic started the “Advent of Claude” campaign, posting a Claude Code tip each day for 31 days.

Claude Code is a highly customizable, automatable, and even programmable development environment.

Let AI Quickly Understand Your Project

New users often ask, “How can I make AI understand my codebase quickly?” Previously you had to copy‑paste files; now Claude Code offers a standard solution.

1. /init : Auto‑generate a project brief

Run /init in a new project. Claude reads the repository and creates a CLAUDE.md file containing:

Build and test commands

Directory structure description

Code style and architecture decisions

…

The file is automatically loaded on each Claude start. For large projects you can add a .claude/rules/ directory (see the earlier “Cursor rules” post) and use Front Matter to activate modular context management.

2. Memory Updates: Dynamic memory editing

Instead of editing CLAUDE.md manually, tell Claude “Update CLAUDE.md: always use bun instead of npm”. Claude writes the new knowledge to the memory file without breaking your flow.

3. @ Mentions: Fast context referencing

Use the @ symbol like an @‑mention in chat to reference code: @src/auth.ts: reference a specific file @src/components/: reference an entire directory @mcp:github: enable or disable a tool (MCP Server)

Fuzzy matching lets you jump from “I need context” to “Claude gets context” instantly.

Essential Shortcuts

Ado emphasizes that mastering these shortcuts multiplies development efficiency.

4. ! Prefix: Run shell commands directly

Type ! git status or ! npm test without leaving the terminal. Claude captures the output, can suggest fixes, and continues the conversation.

5. Double Esc : Time‑travel

Press Esc twice to revert to the previous checkpoint, choosing to roll back the dialogue or the code changes.

6. Ctrl + R : Reverse search

Search previous prompts just like shell reverse‑i‑search.

7. Prompt Stashing: Temporarily save prompts

Press Ctrl + S to stash the current prompt; Claude restores it later, eliminating the need to copy drafts.

8. Prompt Suggestions: Smart completion

When grey suggestions appear, press Tab to accept and edit, or Enter to run immediately.

Session Management

Claude Code is a persistent development environment, not a one‑off chat.

9. Continue & Resume: Seamless continuation

If the terminal closes or the computer dies, use claude --continue to instantly restore the last session, or claude --resume to pick from a history list.

10. Named Sessions: Give sessions a name

Manage sessions like Git branches: /rename api-migration to name the current session, /resume api-migration to restore it.

11. Claude Code Remote: Teleport sessions

Start a task on claude.ai/code, then later run claude --teleport <session_id> to pull the cloud session to a local terminal.

12. /export : Export evidence

Run /export to save the entire dialogue—including prompts, replies, and tool results—as a Markdown file for documentation or post‑mortem.

Productivity Gadgets

13. Vim Mode

Enter /vim to enable Vim keybindings; use h j k l to move and ciw to edit words.

14. /statusline : Custom status bar

Show Git branch, model, token usage, context occupancy, etc., at the bottom.

15. /context : Token insight

Run /context to see how much space System Prompt, MCP Server, memory file, and history occupy, helping you trim usage.

16. /stats : Usage statistics

View favorite models, consecutive usage days, and more—now a brag‑worthy metric.

17. /usage : Quota monitoring

Check current rate limits and progress at any time.

Thinking & Planning

18. ultrathink : Deep‑thinking mode

Add ultrathink to a prompt when designing complex caches or refactoring. Claude allocates up to 32 k tokens for internal reasoning, slower but more accurate.

19. Plan Mode (highly recommended)

Press Shift + Tab twice to enter. Claude reads code, analyzes architecture, drafts a plan, but never modifies code until you approve. You stay the architect; Claude executes.

20. Extended Thinking (API)

When calling the API, enable Extended Thinking to see Claude’s step‑by‑step reasoning blocks—useful for debugging complex logic.

Security & Control

21. Sandbox Mode

Define a boundary once with /sandbox. Example: allow file reads and npm test, but block network requests.

22. YOLO Mode

Use --dangerously-skip-permissions to skip all permission prompts. Use with extreme caution.

23. Hooks: Lifecycle callbacks

Attach shell scripts to events like PreToolUse or SubagentStart. For example, intercept any rm -rf command and raise an alert.

Automation & CI/CD

24. Headless Mode

Pass -p to run Claude as a CLI tool in pipelines: git diff | claude -p "Explain these changes".

25. Commands: Reusable prompts

Save frequent prompts as Markdown files and invoke them with a short command, e.g., /daily-standup to generate a daily report.

Browser Integration

26. Claude Code + Chrome

Install the Chrome extension; Claude can click buttons, fill forms, read console errors, and take screenshots—turning “fix bug and verify” into a single command.

Advanced Agents & Extensions

27. Subagents

Claude can spawn multiple sub‑agents, each with its own 200 k context, to work in parallel and then aggregate results.

28. Agent Skills

Package commands and scripts into a Skill folder. Share a deployment workflow once; teammates instantly gain the expert capability.

29. Plugins Marketplace

Install packaged commands, Skills, Hooks, and MCP Server with /plugin install in one click.

30. LSP Integration

Through Language Server Protocol integration, Claude attains IDE‑level code understanding: real‑time errors, go‑to definition, type info.

31. Claude Agent SDK

The core capabilities (Agent Loop, tool management, context handling) are exposed as an SDK, allowing you to build custom agents with just a few dozen lines of code.

Beyond a Tool – A Philosophy

After reviewing the 31 tricks, the design philosophy becomes clear:

Plan Mode respects human decision‑making.

Hooks and Sandbox give users control.

Subagents and Automation shoulder repetitive work.

As Ado Kukic says, “The best developers aren’t those who hand everything to AI, but those who know when to plan, when to deep‑think, and how to set safe boundaries.”