تخطي للذهاب إلى المحتوى

Boris Cherny's Guide to Claude Code | Pro Tips and Tricks

Practical Guide: Getting Started with Claude Code
6 مايو 2026 بواسطة
FM

Boris Cherny introduces Claude Code – a fully agentic AI assistant for coding. Unlike autocomplete tools, Claude Code handles entire features, functions, files, or bugs at once. It works with any IDE, terminal, or remote environment (SSH, tmux) and requires no indexing – your code stays local and private.

Key Tips

These Boris tips for starting Claude Code efficiently

Start with Q&A 

use Claude Code to explore and understand codebases before editing.

Plan-first prompting 

ask Claude to brainstorm and make a plan before writing code.

Iterative feedback loops 

give Claude a way to check its work (unit tests, screenshots) so it can auto-correct.

Context control 

use CLAUDE.md, slash commands, and @mentions to give Claude necessary context.

Safety & flexibility 

tiered permissions allow auto-approval of safe commands while blocking dangerous ones.

Parallel work 

power users run multiple Claude Code sessions across repos or git worktrees.


Practical Guide: Getting Started with Claude Code

Share this guide 

1. Installation & Setup

  • Requires Node.js. Install with:

    bash

    npm install -g @anthropic-ai/claude-code

  • Run claude terminal-setup for better UX (shift+enter for newlines).

  • Run claude theme to set light/dark mode.

  • Run claude install github-app to enable @mentioning Claude on GitHub issues/PRs.


2. First Steps: Codebase Q&A (No editing yet!)

Use Claude Code to ask questions about your project. Examples:

  • “How is this class used? Show me examples.”

  • “Why does this function have 15 arguments? Check git history.”

  • “What did I ship this week?” (looks at commits by your username)

  • “Fetch GitHub issue #123 and explain the context.”

This trains you and your team on prompting and builds trust in the tool.

3. Basic Editing Workflow

Prompt pattern – Plan first:

text

Before you write code, make a plan for [feature/bug fix]. Run it by me for approval.

Auto-commit & PR:

text

Commit and push. Then make a pull request.

Claude automatically checks git log, creates a branch, commits, pushes, and opens a PR.

4. Feedback Loop for Better Results

Give Claude a way to see its own output so it can iterate. Examples:

  • “Build this web UI. Use Puppeteer to screenshot the result. Keep iterating until it matches the mock.”

  • “Write unit tests. Run them and fix failures.”

5. Providing Context (Critical for Performance)

MethodPurposeScope

CLAUDE.md (project root)

Shared team context: common commands, style guide, core files

Auto-loaded every session

CLAUDE.local.md

Personal overrides, not checked into git

Auto-loaded

Nested CLAUDE.md

Directory-specific instructions

Loaded when Claude works in that dir

/commands (.claude/commands/)

Custom slash commands for automation

Team or personal

@mentions

@file.ts or @folder/

Pull on demand

Enterprise policies

Global allow/block lists for all employees

Managed centrally

Tip: Use /memory to see which context files are being loaded. Use # remember ... to have Claude automatically add something to CLAUDE.md.

6. Keyboard Shortcuts (Terminal)

ActionShortcut
Auto-accept editsShift+Tab (bash still asks)
Stop Claude safelyEscape
Jump back in historyEscape twice
Run bash command (visible to Claude)! command
Remember something# remember ...
Show Claude’s full outputCtrl+R
Resume previous sessionclaude --continue

7. Advanced: SDK & Automation

Use claude -p for non-interactive scripting (like a super intelligent Unix utility):

bash

# Pipe in context, get JSON
git diff | claude -p "Write a commit message" --output-format json

# Fetch log and ask for analysis
cat huge.log | claude -p "Find the most interesting errors"

Use in CI, incident response, or any pipeline.

8. Power User: Parallel Sessions

  • Run multiple Claude Code instances in different terminals.

  • Use git worktrees to have multiple checkouts of the same repo.

  • Combine with SSH/tmux for remote work.

Recommended Adoption Path for Teams

  • Day 1 – Everyone installs Claude Code and does only Q&A (no editing).

  • Day 2 – Add a shared CLAUDE.md with common commands and style guide.

  • Day 3 – Start with plan-first editing on small features.

  • Week 2 – Integrate feedback tools (tests, screenshots) to enable iteration.

  • Ongoing – Use /memory to audit and refine context. Add MCP servers or custom slash commands for team-specific tools.