#cursor#cli#agent#plan#ask

Cursor CLI Three Modes: Agent / Plan / Ask

3 min read

Three modes, one sentence to tell them apart: who can touch your code, and who can only talk. Once you get that, picking the right mode is easy.

One-Sentence Comparison

Mode One Sentence Does it modify your files?
Agent "Do" something for me ✅ Yes
Plan "Plan" how to do it first, then I'll decide Not yet — only after you approve
Ask Just "ask" a question or look something up ❌ No

Think of them as three colleagues: Agent is the one who says "hand it to me, I'll just do it"; Plan is the one who says "let me write up a proposal for you to review first"; Ask is the one who says "I'll only answer, I won't touch the files" — pick the right person and save yourself half the trouble.

Agent Mode (Default)

Use case: Have the AI directly edit code, add features, or refactor. You want results, not a discussion.

cursor agent "Change login to use JWT"
  • Reads files, edits files, may run commands
  • You can accept or reject each step — you're in control

Plan Mode

Use case: Not sure how to approach something, or worried about the AI making too many changes at once? See the plan before giving the go-ahead.

cursor agent --plan "Refactor the entire API layer"
  • The AI will ask a few questions and propose an approach
  • It only acts after you agree — no surprises

When to use: For large changes, or whenever you want to "see it clearly before committing."

Ask Mode (Read-Only)

Use case: You just want to ask "what does this code do" or "is there a better way to write this" — and you absolutely do not want any files changed.

cursor ask "What is the logic in src/auth.ts?"
  • Only reads, analyzes, and answers — never writes
  • Great for CI, or environments where you only want to inspect, not modify

Switching Modes Mid-Conversation

Already inside a cursor agent conversation? Use slash commands to switch instantly:

  • /plan → Switch to Plan mode (plan first, act later)
  • /ask → Switch to Ask mode (read-only, ask freely)

No need to exit and restart — it's like switching gears on the fly.

Summary

  • Agent: gets things done, modifies files — use this when you want results
  • Plan: plans first, acts only after you approve — use this when you want safety
  • Ask: questions only, read-only — use this when you want answers without changes

Next: 04-non-interactive — Non-interactive mode for scripts and CI, let the AI run unattended