Agent mode in the IDE
Edits multi-file projects, runs terminal commands, and iterates with your repo as context—not a one-off chat pane.
Cursor Agent runs inside the editor with tools, terminal, and MCP. You extend it with Skills—`SKILL.md` instructions and scripts—for repeatable workflows without rebuilding the agent each time.
Open CursorDownload from cursor.com, sign in, and open your repository. Enable Agent in the chat/composer UI per current docs.
Configure project rules and any MCP servers your stack needs (GitHub, Linear, browser, etc.).
Add a `SKILL.md` under `.cursor/skills/` or your user skills folder; follow Cursor’s skill format for name, description, and steps.
Start with read-only exploration, then enable terminal writes once you trust approvals and `.cursorignore` boundaries.
Give the agent a clear definition of done, files in scope, and commands it should run to verify.
When a workflow is packaged, name the skill or let the agent auto-select based on the skill description field.
Treat Agent like a junior teammate—scan patches and terminal actions before merge.
When the agent misses a step, update SKILL.md once so the next run is consistent.
Edits multi-file projects, runs terminal commands, and iterates with your repo as context—not a one-off chat pane.
Pack domain workflows (deploy, review, migrations) as skills the agent loads when relevant or when you invoke them.
Executes build/test/lint commands with approval gates; pairs with MCP for APIs, docs, and issue trackers.
Combine `.cursor/rules`, AGENTS.md, and skills so the whole team shares the same agent behavior.
Agent implements a ticket, runs tests, and opens a PR following your skill-defined checklist.
Use a skill to enforce migration patterns across many files with consistent verification steps.
MCP pulls logs or tickets; a skill documents how to reproduce, patch, and roll out safely.
Cursor Agent is an AI coding agent inside the Cursor IDE. It edits multiple files, runs terminal commands with approval, and uses your repository as context—not a single-file chat snippet.
Skills are folders with a SKILL.md file that describe when to trigger a workflow and which steps to follow. Install them in project or user skill directories so Agent reuses the same release or review checklist.
Treat it like a junior teammate: review diffs and terminal commands, use branch protection and CI, start with read-only exploration, and tighten `.cursorignore` before enabling destructive commands.
Note: Cursor Agent, Skills, and MCP settings change frequently—confirm in Cursor Docs before changing production repos.