>_ restart prompt
Write a clean restart prompt
Your agent keeps going in circles. Answer four questions and get a prompt for a fresh chat that says what works, what is broken and what not to try again.
Nothing you type leaves this browser. There is no send button and nothing is saved.
What you were trying to get working, in a sentence or two.
Anything the new chat must leave alone.
What you see, and any error message word for word.
Fixes the agent tried that did not work.
First, start clean in Cursor
- If the last edits broke things, use Restore checkpoint on the last message that worked.
- Open a new chat with the + button at the top of the chat panel, then paste the prompt.
First, start clean in Claude Code
- If you need the code back, type /rewind (or press Esc twice) and pick the last point that worked. Files changed by shell commands are not restored.
- Type /clear to start a fresh conversation, then paste the prompt.
First, start clean in Lovable
- Open History, pick the last version that worked and choose Revert. Data changes made since then are not undone.
- Paste the prompt as your next message.
First, start clean in Replit
- Roll back to the last checkpoint that worked. The database is only rolled back if you tick it.
- Start a new Agent chat and paste the prompt.
First, start clean in Bolt.new
- Restore the last version that worked from the version history.
- Paste the prompt as your next message.
First, start clean in v0
- Open the last version that worked and continue from it.
- Paste the prompt as your next message.
First, start clean in Base44
- Go back to the last version that worked from the version history.
- Start a new chat and paste the prompt.
First, start clean in Codex
- If files changed, undo them first (for example with Git).
- Type /clear to start a fresh chat, then paste the prompt.
First, start clean
- Undo back to the last version that worked, with your tool's history or with Git.
- Open a brand new chat and paste the prompt.
Fill in what should happen and what happens instead, and your prompt appears here.
Why a fresh chat works
- A long chat keeps every failed attempt in view, and the agent keeps building on them.
- A new chat with a short, exact brief starts from what you know, not from what went wrong.
- Saying what not to try again stops the same fix coming back in new words.