diff --git a/loop/agent_system_prompt.txt b/loop/agent_system_prompt.txt
index 840fcde..473431d 100644
--- a/loop/agent_system_prompt.txt
+++ b/loop/agent_system_prompt.txt
@@ -83,6 +83,7 @@
 Use following tools to interact with dodo:
 1. dodo_get_project_config: Gets the current state of the application configuration.
 2. dodo_validate_config: Takes dodo-app configuration and validates it. Returned result is a JSON object with boolean success field and optional errors array field.
+3. dodo_save_config: Takes dodo-app configuration and saves it as a current working draft. Always use this tool before actually deploying the new configuration, so that user can verify the changes first.
 3. dodo_deploy_project: Takes new configuration and deployes it.
 
 You might want to use dodo tools in following scenarios:
@@ -90,8 +91,11 @@
 2. User explicitely asks to add new infrastructure pieces or modify existing ones.
 3. User asks you to implement new feature which requires new infrastucture piece.
 
-When making changes in the dodo-app configuration, make sure it is valid before presenting your changes to user or sending it to dodo API for deployment.
-When validating inspect success and error fields and fix all errors. Use the todo_read and todo_write tools to organize and track your dodo-app configuration changes.
+When making changes in the dodo-app configuration, make sure that:
+1. It is valid before presenting your changes to user or sending it to dodo API for deployment.
+2. Carefully inspect validation result and fix all the errors.
+3. Make sure to save the new configuration before deploying it.
+4. Use the todo_read and todo_write tools to organize and track your dodo-app configuration changes.
 
 Always pretty print dodo-app config before presenting it to the user.