RDBC-998 Write tests for Python AI Agents

[poissoncorp]

RDBC-998 Write tests for Python AI Agents

Changes

ravendb/documents/operations/ai/agents/__init__.py

• Exported AiAgentToolQueryOptions in the public API.

ravendb/tests/ai_agent_tests/test_ai_agents.py — 15 integration tests

CRUD operations against the real embedded test server.

Test	What it covers
test_create_and_get_agent	Create an agent, fetch it back, verify all fields round-trip
test_update_agent	Update system prompt and verify the change persists
test_delete_agent	Delete an agent, confirm it is gone
test_agent_with_query_tool	Agent with a query tool; verify tool fields
test_agent_with_action_tool	Agent with an action tool; verify tool fields
test_agent_with_multiple_tools	Agent with both query and action tools
test_agent_with_parameters	Agent with typed parameters and sample object
test_agent_connection_string_name	Connection string name is stored and returned correctly
test_agent_system_prompt	System prompt is stored and returned correctly
test_agent_identifier	Agent identifier is stored and returned correctly
test_list_agents	List all agents; verify the created agent appears
test_agent_with_query_tool_options_add_to_initial_context	AiAgentToolQueryOptions.add_to_initial_context persists
test_agent_with_query_tool_options_allow_model_queries	AiAgentToolQueryOptions.allow_model_queries persists
test_create_multiple_agents	Create several agents; all appear in the list
test_delete_nonexistent_agent	Deleting a non-existent agent raises an appropriate error

ravendb/tests/ai_agent_tests/test_ai_agents_conversation_mock.py — 21 hybrid tests

Agent CRUD runs against the real server; store.maintenance.send is patched so
RunConversationOperation calls return crafted ConversationResult objects — no
LLM API key required.

Group	Test	What it covers
Basic flow	test_basic_conversation_returns_answer	run() returns the mocked answer
	test_conversation_id_and_change_vector_are_stored	_conversation_id / _change_vector are updated after a call
	test_usage_is_populated	AiUsage fields (prompt/completion/total tokens) are propagated
	test_elapsed_is_populated	Elapsed time is set on the result
	test_context_manager_usage	AiConversation works as a context manager
Action handlers	test_handle_invokes_handler_and_sends_response	handle() callback is called with parsed args; response is sent back
	test_receive_invokes_handler_with_request_and_args	receive() callback receives both the raw request and parsed args
	test_multi_turn_action_loop	run() loops correctly across multiple ACTION_REQUIRED turns until DONE
Error strategies	test_handler_error_send_to_model	Handler exception with SEND_ERRORS_TO_MODEL — conversation continues
	test_handler_error_raise_immediately	Handler exception with RAISE_IMMEDIATELY — exception propagates
Unhandled actions	test_on_unhandled_action_is_called	on_unhandled_action callback fires; conversation continues after fallback response
	test_no_handler_raises_runtime_error	No handler and no on_unhandled_action → RuntimeError
Artificial actions	test_add_artificial_action_with_response	add_artificial_action_with_response() injects the action into the operation
	test_add_artificial_action_validates_tool_id	Empty/blank tool ID raises ValueError
	test_add_artificial_action_validates_response_not_none	None response raises ValueError
Creation options	test_conversation_with_creation_options	AiConversationCreationOptions (parameters, expiration) are forwarded to the operation
Streaming	test_stream_collects_chunks_and_returns_answer	stream() completes and returns a DONE result
Validation	test_set_user_prompt_empty_raises	Empty/blank prompt raises ValueError
	test_required_actions_before_run_raises	Accessing required_actions before run() raises RuntimeError
	test_required_actions_after_run_returns_list	required_actions returns an empty list after a DONE result
	test_duplicate_action_handler_raises	Registering two handlers for the same action name raises ValueError