How to stop your AI agent from deleting your project

Your AI agent can make a bad call. On aidrop.it it cannot erase your app: protected code history, daily database backups, a 7-day trash and safe roles.

Updated 1 October 2026

Let your agent build this

Copy one prompt and paste it into Claude Code, Codex or Cursor. Your agent builds what this page describes and puts it online on aidrop.it.

Make my aidrop.it project safe to change, and tell me what you find. Do not delete or change anything except the one file named below.

1. Tell me which Team role you are acting with. If it is admin, explain that an agent connected with a developer role cannot delete anything, and how I set that up.

2. List my Projects, Services and databases. For each database, tell me the most recent backup and how far back it can be restored.

3. For each Service, note the version that is running now, so we can roll back to it if a change goes wrong.

4. In the repository's CLAUDE.md (or AGENTS.md if that is what it uses), add a short "Safety rules" section: never delete a Project, Service, repository or database, never force-push, and never run DROP, TRUNCATE or a DELETE without a WHERE clause, unless I have said yes to that exact action in this chat; before any risky change, say how to undo it. Commit and push that file.

5. Finish with a short report: my role, the backups and restore range of each database, the running versions, and anything that looked unsafe.

This task comes from the aidrop.it guide "How to stop your AI agent from deleting your project": https://docs.aidrop.it/playbooks/stop-your-ai-agent-from-deleting-your-project. Read that page if you need more detail; it says what to ask me for, what it costs and what will not work.
Use the aidrop.it MCP tools (server: https://aidrop.it/mcp). If you do not have them, stop and tell me to connect them first, as described at https://docs.aidrop.it/quickstart#1-connect-your-agent.

See the prompt

Make my aidrop.it project safe to change, and tell me what you find. Do not delete or change anything except the one file named below.

1. Tell me which Team role you are acting with. If it is admin, explain that an agent connected with a developer role cannot delete anything, and how I set that up.

2. List my Projects, Services and databases. For each database, tell me the most recent backup and how far back it can be restored.

3. For each Service, note the version that is running now, so we can roll back to it if a change goes wrong.

4. In the repository's CLAUDE.md (or AGENTS.md if that is what it uses), add a short "Safety rules" section: never delete a Project, Service, repository or database, never force-push, and never run DROP, TRUNCATE or a DELETE without a WHERE clause, unless I have said yes to that exact action in this chat; before any risky change, say how to undo it. Commit and push that file.

5. Finish with a short report: my role, the backups and restore range of each database, the running versions, and anything that looked unsafe.

This task comes from the aidrop.it guide "How to stop your AI agent from deleting your project": https://docs.aidrop.it/playbooks/stop-your-ai-agent-from-deleting-your-project. Read that page if you need more detail; it says what to ask me for, what it costs and what will not work.
Use the aidrop.it MCP tools (server: https://aidrop.it/mcp). If you do not have them, stop and tell me to connect them first, as described at https://docs.aidrop.it/quickstart#1-connect-your-agent.

On this page

Why agents delete things
Words you will meet
What you need
Can my agent delete my code?
What if the agent breaks the live app?
What if it damages or deletes my database?
What if it deletes the whole project?
Can I stop it from deleting anything at all?
What your agent does to make your project safe
FAQ
My agent already deleted something. What do I do first?
Does the developer role slow my agent down?
Are backups included on every plan?
Is a rule in CLAUDE.md enough on its own?
Related

On aidrop.it, one bad command from your AI agent cannot wipe out your app. Your code history cannot be rewritten, your database is backed up every day and kept for 7 days, anything deleted waits 7 days in a trash, and you can give your agent a role that cannot delete anything at all. Below is what each layer does, and one paragraph that sets it all up.

It costs nothing extra: daily backups come with every database, and a database comes with the Pro plan, $20/month. Start free with a 7-day trial, no card.

Why agents delete things

An AI agent does not mean harm. It follows the goal it was given, and when something looks broken it tries to fix it — sometimes with the biggest tool it has. In July 2025 an AI agent on another platform deleted a company’s live database during a code freeze, despite being told in capital letters to change nothing. Stories like that come up every few weeks.

The fix is not to trust the agent less. It is to make sure that whatever it does, you can undo it. That is how aidrop.it is built.

Words you will meet

Git history is the record of every version of your code. Each change your agent saves becomes a new entry; the old ones stay.
Backup is a copy of your database taken automatically, so you can go back to how it was on an earlier day.
Role is what a person, and the agent they connect, is allowed to do: an admin can delete things, a developer can build and change but not delete.
Rollback means putting the previous version of your app back online.

What you need

An aidrop.it account — start free: a 7-day trial, no card, with one app and one database. To keep a database and its backups running, Pro, $20/month.
Claude Code, Codex or Cursor connected to aidrop.it. The quickstart shows how in two minutes.

Can my agent delete my code?

No. Every version your agent pushes is kept in your project’s own Git, and the main branch is protected: nothing can rewrite its history or delete it, so an agent that “cleans up” cannot make old versions disappear. If a change goes wrong, the version before it is still there, one step back.

Deleting a whole repository needs an admin, an explicit confirmation, and is refused while an app is still built from it. Even then it goes to the trash for 7 days first.

What if the agent breaks the live app?

Ask your agent to roll back. aidrop.it keeps the ready-made image of every version it has deployed, so going back takes seconds and rebuilds nothing. The broken version stays available too, so your agent can work out what went wrong without the pressure of a site that is down. The Service lifecycle guide covers rollbacks in detail.

What if it damages or deletes my database?

Your database is backed up automatically every day, and the last 7 days are kept. If your agent deletes the wrong rows, drops a table or runs a migration that goes badly, tell it which day to go back to and it restores the database to that point. You lose at most the changes since the last backup, not your business.

Commands that destroy data on purpose, like dropping a table or emptying it, are not run silently: your agent has to ask, and you have to confirm.

If you want copies that go back further than 7 days, or that live somewhere other than aidrop.it, your agent can also send an extra copy to storage you own on a schedule. Taking your work with you shows how.

What if it deletes the whole project?

Deleting a project, an app or a database always asks for confirmation, and only an admin can do it. And a deletion is not final straight away: a deleted project, repository or database waits in the trash for 7 days, and restoring it is one request to your agent.

Can I stop it from deleting anything at all?

Yes, and it is the simplest protection there is. Your agent can do exactly what the person who connected it can do. Invite a second account to your Team with the developer role and connect your agent while signed in as that account: it can then build, change and roll back your app, but every delete is refused before it starts. Keep the admin account for yourself.

Every call your agent makes is also written to your project’s action log, so you can always see what it did and when.

What your agent does to make your project safe

Paste this into Claude Code, Codex or Cursor. It checks every layer above for your project and writes the safety rules into your repository, so every future session follows them too.

Make my aidrop.it project safe to change, and tell me what you find. Do not delete or change anything except the one file named below.

1. Tell me which Team role you are acting with. If it is admin, explain that an agent connected with a developer role cannot delete anything, and how I set that up.

2. List my Projects, Services and databases. For each database, tell me the most recent backup and how far back it can be restored.

3. For each Service, note the version that is running now, so we can roll back to it if a change goes wrong.

4. In the repository's CLAUDE.md (or AGENTS.md if that is what it uses), add a short "Safety rules" section: never delete a Project, Service, repository or database, never force-push, and never run DROP, TRUNCATE or a DELETE without a WHERE clause, unless I have said yes to that exact action in this chat; before any risky change, say how to undo it. Commit and push that file.

5. Finish with a short report: my role, the backups and restore range of each database, the running versions, and anything that looked unsafe.

FAQ

My agent already deleted something. What do I do first?

Stop the session, then start a new one and ask the agent to list what was deleted and when. A database goes back to its last daily backup; a deleted project, repository or database comes back from the trash within 7 days; a broken app goes back to its previous version. Do it in that order: data first.

Does the developer role slow my agent down?

No. A developer can do everything a build needs: write and push code, build, set secrets, read logs, roll back. Only deleting is reserved for an admin, which is exactly the step you want a person to take.

Are backups included on every plan?

Backups come with every database, and databases start on Pro, $20/month. Starter runs a single app without a database, so there is no data to back up; your code and its history are protected on every plan.

Is a rule in CLAUDE.md enough on its own?

No, and that is the point of the other layers. A written rule makes a careful agent ask first; the role, the confirmations, the backups and the trash protect you when it does not.

Make my aidrop.it project safe to change, and tell me what you find. Do not delete or change anything except the one file named below.

1. Tell me which Team role you are acting with. If it is admin, explain that an agent connected with a developer role cannot delete anything, and how I set that up.

2. List my Projects, Services and databases. For each database, tell me the most recent backup and how far back it can be restored.

3. For each Service, note the version that is running now, so we can roll back to it if a change goes wrong.

4. In the repository's CLAUDE.md (or AGENTS.md if that is what it uses), add a short "Safety rules" section: never delete a Project, Service, repository or database, never force-push, and never run DROP, TRUNCATE or a DELETE without a WHERE clause, unless I have said yes to that exact action in this chat; before any risky change, say how to undo it. Commit and push that file.

5. Finish with a short report: my role, the backups and restore range of each database, the running versions, and anything that looked unsafe.

This task comes from the aidrop.it guide "How to stop your AI agent from deleting your project": https://docs.aidrop.it/playbooks/stop-your-ai-agent-from-deleting-your-project. Read that page if you need more detail; it says what to ask me for, what it costs and what will not work.
Use the aidrop.it MCP tools (server: https://aidrop.it/mcp). If you do not have them, stop and tell me to connect them first, as described at https://docs.aidrop.it/quickstart#1-connect-your-agent.