AI toolsο
Claude Codeο
This is a tool which interfaces between your command line and the LLM Claude uses.
Commandsο
Commands are run but using /<command>. Can see a list by just typing the /
init: Claude analyses your code and makes a summary in theclaude.mdfile for future contextcompact: summarises the chat for concise contextclear: refresh the context from the chat
You can also implement custom commands in .claude/commands/<name.md>, including with arguments with the @ symbol
Shortcutsο
#: Memory mode, give instruction of something to commit to the project memory@<filepath>: point to a specific file for context. Can also be in theCLAUDE.mdfile, where the file is pasted in place so always part of the context promptescape: interrupt claude
double escape: go back in the chat history to remove some context that is no longer needed
Instruction Filesο
You can have instruction files in 3 different places:
Root of the repo: in version control and applies to all devs
claude.local.md: personalised instructions not in version control for the project~/.claude/CLAUDE.md: applies to all projects on your machine
MCP serverο
A gateway for accessing external tools, for example βplaywrightβ for interacting with browsers.
It is an open standard for interacting with AI tools, so you can also write your own.
Hooksο
Allows you to run a command before or after a tool runs. For example a tool might be one to read a file to give that to the claude LLM.
You could run a custom hook before this tool is run, perhaps to check the file is allowed to be sent to the LLM. One after might apply code formatting.
Hooks can be defined again in the repo, both personal and repo wide, and globally on your machine.
Other hook stages also exist, such as session start and end, and userpromt submitted.
Miscο
Can instruct claude to write commit messages and commit files
claude SDK: SDK available in python which lets you programmatically invoke claude code commands. However, not all commands are available as using the CLI.