PrimAITE contains a selection of possible actions that can be exercised within a training environment. Actions provided as a part of PrimAITE can be seen within `src/primaite/game/agent/actions`. `Note`: Agents are only able to perform the actions listed within it's action_map, defined within it's configuration YAML. See :ref:`custom_environment` for more information.
Actions within PrimAITE follow a default format, as seen below and in ``actions.py``. It's important that they have an identifier when declared, as this is used when creating the training environment.
Any custom actions should then be added to the `ActionManager` class, and the `act_class_identifiers` dictionary. This will map the action class to the corresponding action type string that would be passed through the PrimAITE `request_system`.
Where an action would cause a request to be sent through the PrimAITE RequestManager, a `form_request` method is expected to be defined within the Action Class. This should format the action into a format that can be ingested by the `RequestManager`. Examples of this include the `NodeFolderCreateAction`, which sends a formed request to create a folder on a given node (seen below):
Agents which use the `ProxyAgent` class within PrimAITE are able to use Action Masking. This allows the agent to know if the actions are valid/invalid based on the current environment.
Information on how to ensure this can be applied to your custom action can be found in :ref:`action_masking`