Claude Code Cheatsheet

Some files, features, and shortcuts to keep in mind when using Claude Code (from Anthropic's Course 'Claude Code: A Highly Agentic Coding Assitant').

Some New Learnings:

For searching within codebase, Claude Code uses agentic search (instead of indexing) so it doesn’t send your code to outside servers
May want to switch model to Opus for more complex tasks

New features I learned about:

Screenshots:

just drag and drop image files

Hooks:

add in code/operations at any point in coding lifecycle — before/after tool execution, when user submits a prompt, right before claude concludes it prompts

e.g. write “ALL DONE” after Claude is done with a tool call.
e.g. run unit tests after all the code has been written.

Github integration

Can tag Claude in issues and PR comments
Claude can do a code review on new PRs

Parallel feature implementation

Can have Claude work on multiple features simultaneously (by using git worktrees and custom implement feature commands)
Claude can also fix merge conflicts and create PRs with all features.
Need to create a custom command in commands/implement-feature.md file to specify what needs to be done for feature implementation

Jupyter notebook Compatible!

Files to be aware of:

Claude.md — for project, shared with other engineers
Claude.local.md — personal custom instructions/customizations for claude
~/.claude/CLAUDE.md — global for all projects on machine
settings.local.json — specifies which commands require which permissions + where hooks live

Helpful Commands:

— add something to memory

/ide

— so claude has access to the ide you’re using and you can give claude context to files in your ide

— adding context to a certain file

/clear

— to clear a conversation and have a new start (refreshes context window)

/compact

— clear history but maintain summary of old chat

/mcp

— connect to an mcp server

/resume

— to return to an old conversation you were having

/install-github-app

— installs github integration!

MCP servers:

install server + use with /mcp command.

playwright: can use this to have AI verify that UI/UX features were implemented correctly (equivalent to running app + manually checking visually after a change)
figma make (figma-dev-mode-mcp-server): can take a figma mockup and build underlying code in nextjs app