Kenhuang Academy
  • English

      • Introduction

      What is Cursor?

      Cursor is an AI-native code editor / coding agent from Anysphere. It grew out of the VS Code ecosystem (extensions, keybindings, and settings mostly migrate), but the official docs now emphasize using Agent to understand a repo, plan features, edit code, run checks, and review diffs—not just “an editor with autocomplete.”

      Authoritative entry points:

      The product has three surfaces: the desktop editor, the terminal Agent CLI (agent), and Cloud Agents you can start from the browser, Slack, GitHub, and similar tools. This tutorial follows the desktop Agent; CLI and Cloud appear where they matter.

      Three components (mental model)

      Officially, an agent is three pieces:

      PieceMeaningWhat you control
      InstructionsSystem prompt + rules.cursor/rules, user rules, Team Rules, AGENTS.md
      ToolsSearch, read, edit, shell, web, browser, image gen, …Approve commands, toggle MCP, pick a Run Mode
      ModelThe model for this taskAuto or a named model; trust the in-app list
      flowchart LR
  U[Your goal] --> I[Instructions]
  U --> M[Model]
  I --> A[Agent loop]
  M --> A
  A --> T[Tools]
  T --> A
  A --> D[Diff / checkpoints]

      Cursor tunes instructions and tools per model. When you switch models, you keep the same rules and MCP—you do not rewrite a prompt stack.

      Versus VS Code + Copilot

      DimensionCursorVS Code + Copilot
      ShapeStandalone editor + AgentHost editor + extension
      Daily completeTab (multi-line, cross-file, jump-in-file)Copilot / Copilot Chat
      Multi-file editsSidepane Agent (Ctrl+I / Cmd+I)Chat / Edits, plan-dependent
      Repo-scale planningPlan Mode researches, then waitsMostly conversational
      ExtensibilityVS Code extensions + Cursor Marketplace (Skills / MCP / plugins)VS Code Marketplace
      Rules.cursor/rules, AGENTS.md, Team RulesInstruction files vary by plugin

      Pick: Stay if you are locked into VS Code and only need inline complete. Choose Cursor if you want planning, multi-file edits, tests, and MCP in one window. First launch can import VS Code settings and extensions, so migration is usually cheap.

      Versus Claude Code and Codex

      This site has full courses: Claude Code and Codex. All three are coding agents; the door you walk through differs.

      DimensionCursorClaude CodeCodex
      Primary doorDesktop IDE (editor + sidepane Agent)Terminal CLI (claude)Desktop app / CLI / IDE extension
      StrengthDiffs, files, interactive edit loopLong autonomous terminal work, permissions, hooksApp + Cloud + automation
      Rules.cursor/rules, AGENTS.mdCLAUDE.md, SkillsAGENTS.md, Skills
      MCP.cursor/mcp.json or ~/.cursor/mcp.jsonProject / user MCPMCP config and plugin marketplace
      Best whenYou want to watch code while reviewing AgentLong scripts, CI, keyboard-onlyTeams already on OpenAI

      Recommendation (expanded later): use Cursor for the interactive edit loop; use Claude Code / Codex for long, autonomous terminal jobs. Write shared rule files so the three stacks do not invent conflicting “oral rules.”

      When it fits

      Good fit:

      • Unfamiliar repos: ask Agent for entry points and must-read files
      • Multi-file features and refactors that need a plan before code
      • Browser checks for UI, or MCP for issues / designs / docs
      • Teams that share one rules + MCP set

      Be careful:

      • Sensitive repos: confirm privacy mode and Team / Enterprise data policy
      • Untrusted MCP: a server can call APIs and run local commands for you
      • Treating checkpoints as Git: restore reverts Agent-touched files, not the whole history

      How to read this course

      Finish Installation and Quick Start, then Agent. Tab and Ctrl+K are still the highest-frequency actions every day—even though Agent is the headline, do not skip Tab and Inline Edit.

      Next

      评论