Here are few things we're working on next around safety:
- Dynamic permissions for agents: agents get API keys with limited scope. Whenever they need to expand the permission, they will ask you for approval. And the expanded scope only applies to the CURRENT task.
- Reversible: Git for backend. All the write operations will be snapshotted, so if agents make terrible mistakes, you can always roll back.
- Dynamic permissions for agents: agents get API keys with limited scope. Whenever they need to expand the permission, they will ask you for approval. And the expanded scope only applies to the CURRENT task.
- Reversible: Git for backend. All the write operations will be snapshotted, so if agents make terrible mistakes, you can always roll back.