How to better use claude code? (中文版本)

prompt should clear , detailed through the lession not only should we learn,but also work the auth of cc action will be stored in settings.local.json

explain

if the current context is over,plz turn off and start another new context MCP server - add as needed ,do not overdo it CLAUDE.md - keep it clean , only put the necessary information, you can also put your rules of project in there

location

put your folder of .claude in the root dir of project dir if you put folder of .claude if not in the root of project ,you can put .claude in the ~/.claude ,that has global scope

CLAUDE.md

-say what does this project work? -technical stack -variable name,style of code -rules and banned

commands

type less words to do more job /review -- "check out the commit quality of code,and gave feedback,if has bug then fix it" only one command to replace template of prompt .claude/commands/command_name.md

mcp server

if function calling let llm could use local function then mcp server let llm could use local server through model context protocol like the playwright server , llm -> mcp context protocol(expose the server) -> local mcp server ->llm control browser

claude mcp add playwright -- npx @playwright/mcp@latest
claude mcp add github -e GITHUB_PERSONAL_ACCESS_TOKEN=token -- npx -y @modelcontextprotocol/server-github

skills

natural language to execute standard operation process