diff --git a/tests/test_anthropic_adapter.py b/tests/agent/test_anthropic_adapter.py
similarity index 100%
rename from tests/test_anthropic_adapter.py
rename to tests/agent/test_anthropic_adapter.py
diff --git a/tests/test_auxiliary_config_bridge.py b/tests/agent/test_auxiliary_config_bridge.py
similarity index 98%
rename from tests/test_auxiliary_config_bridge.py
rename to tests/agent/test_auxiliary_config_bridge.py
index 0151daf2a1..91dea15af6 100644
--- a/tests/test_auxiliary_config_bridge.py
+++ b/tests/agent/test_auxiliary_config_bridge.py
@@ -13,7 +13,7 @@ from unittest.mock import patch, MagicMock
 import pytest
 import yaml
 
-sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", ".."))
 
 
 def _run_auxiliary_bridge(config_dict, monkeypatch):
@@ -199,7 +199,7 @@ class TestGatewayBridgeCodeParity:
 
     def test_gateway_has_auxiliary_bridge(self):
         """The gateway config bridge must include auxiliary.* bridging."""
-        gateway_path = Path(__file__).parent.parent / "gateway" / "run.py"
+        gateway_path = Path(__file__).parent.parent.parent / "gateway" / "run.py"
         content = gateway_path.read_text()
         # Check for key patterns that indicate the bridge is present
         assert "AUXILIARY_VISION_PROVIDER" in content
@@ -213,7 +213,7 @@ class TestGatewayBridgeCodeParity:
 
     def test_gateway_no_compression_env_bridge(self):
         """Gateway should NOT bridge compression config to env vars (config-only)."""
-        gateway_path = Path(__file__).parent.parent / "gateway" / "run.py"
+        gateway_path = Path(__file__).parent.parent.parent / "gateway" / "run.py"
         content = gateway_path.read_text()
         assert "CONTEXT_COMPRESSION_PROVIDER" not in content
         assert "CONTEXT_COMPRESSION_MODEL" not in content
diff --git a/tests/test_context_references.py b/tests/agent/test_context_references.py
similarity index 100%
rename from tests/test_context_references.py
rename to tests/agent/test_context_references.py
diff --git a/tests/test_credential_pool.py b/tests/agent/test_credential_pool.py
similarity index 100%
rename from tests/test_credential_pool.py
rename to tests/agent/test_credential_pool.py
diff --git a/tests/test_credential_pool_routing.py b/tests/agent/test_credential_pool_routing.py
similarity index 100%
rename from tests/test_credential_pool_routing.py
rename to tests/agent/test_credential_pool_routing.py
diff --git a/tests/test_crossloop_client_cache.py b/tests/agent/test_crossloop_client_cache.py
similarity index 100%
rename from tests/test_crossloop_client_cache.py
rename to tests/agent/test_crossloop_client_cache.py
diff --git a/tests/test_display.py b/tests/agent/test_display.py
similarity index 100%
rename from tests/test_display.py
rename to tests/agent/test_display.py
diff --git a/tests/test_insights.py b/tests/agent/test_insights.py
similarity index 100%
rename from tests/test_insights.py
rename to tests/agent/test_insights.py
diff --git a/tests/test_model_metadata_local_ctx.py b/tests/agent/test_model_metadata_local_ctx.py
similarity index 100%
rename from tests/test_model_metadata_local_ctx.py
rename to tests/agent/test_model_metadata_local_ctx.py
diff --git a/tests/cli/__init__.py b/tests/cli/__init__.py
new file mode 100644
index 0000000000..e69de29bb2
diff --git a/tests/test_branch_command.py b/tests/cli/test_branch_command.py
similarity index 100%
rename from tests/test_branch_command.py
rename to tests/cli/test_branch_command.py
diff --git a/tests/test_cli_approval_ui.py b/tests/cli/test_cli_approval_ui.py
similarity index 100%
rename from tests/test_cli_approval_ui.py
rename to tests/cli/test_cli_approval_ui.py
diff --git a/tests/test_cli_background_tui_refresh.py b/tests/cli/test_cli_background_tui_refresh.py
similarity index 100%
rename from tests/test_cli_background_tui_refresh.py
rename to tests/cli/test_cli_background_tui_refresh.py
diff --git a/tests/test_cli_browser_connect.py b/tests/cli/test_cli_browser_connect.py
similarity index 100%
rename from tests/test_cli_browser_connect.py
rename to tests/cli/test_cli_browser_connect.py
diff --git a/tests/test_cli_context_warning.py b/tests/cli/test_cli_context_warning.py
similarity index 100%
rename from tests/test_cli_context_warning.py
rename to tests/cli/test_cli_context_warning.py
diff --git a/tests/test_cli_extension_hooks.py b/tests/cli/test_cli_extension_hooks.py
similarity index 100%
rename from tests/test_cli_extension_hooks.py
rename to tests/cli/test_cli_extension_hooks.py
diff --git a/tests/test_cli_file_drop.py b/tests/cli/test_cli_file_drop.py
similarity index 100%
rename from tests/test_cli_file_drop.py
rename to tests/cli/test_cli_file_drop.py
diff --git a/tests/test_cli_init.py b/tests/cli/test_cli_init.py
similarity index 100%
rename from tests/test_cli_init.py
rename to tests/cli/test_cli_init.py
diff --git a/tests/test_cli_interrupt_subagent.py b/tests/cli/test_cli_interrupt_subagent.py
similarity index 100%
rename from tests/test_cli_interrupt_subagent.py
rename to tests/cli/test_cli_interrupt_subagent.py
diff --git a/tests/test_cli_loading_indicator.py b/tests/cli/test_cli_loading_indicator.py
similarity index 100%
rename from tests/test_cli_loading_indicator.py
rename to tests/cli/test_cli_loading_indicator.py
diff --git a/tests/test_cli_mcp_config_watch.py b/tests/cli/test_cli_mcp_config_watch.py
similarity index 100%
rename from tests/test_cli_mcp_config_watch.py
rename to tests/cli/test_cli_mcp_config_watch.py
diff --git a/tests/test_cli_new_session.py b/tests/cli/test_cli_new_session.py
similarity index 100%
rename from tests/test_cli_new_session.py
rename to tests/cli/test_cli_new_session.py
diff --git a/tests/test_cli_plan_command.py b/tests/cli/test_cli_plan_command.py
similarity index 100%
rename from tests/test_cli_plan_command.py
rename to tests/cli/test_cli_plan_command.py
diff --git a/tests/test_cli_prefix_matching.py b/tests/cli/test_cli_prefix_matching.py
similarity index 100%
rename from tests/test_cli_prefix_matching.py
rename to tests/cli/test_cli_prefix_matching.py
diff --git a/tests/test_cli_preloaded_skills.py b/tests/cli/test_cli_preloaded_skills.py
similarity index 100%
rename from tests/test_cli_preloaded_skills.py
rename to tests/cli/test_cli_preloaded_skills.py
diff --git a/tests/test_cli_provider_resolution.py b/tests/cli/test_cli_provider_resolution.py
similarity index 100%
rename from tests/test_cli_provider_resolution.py
rename to tests/cli/test_cli_provider_resolution.py
diff --git a/tests/test_cli_retry.py b/tests/cli/test_cli_retry.py
similarity index 96%
rename from tests/test_cli_retry.py
rename to tests/cli/test_cli_retry.py
index 74e2512bfe..b287b45754 100644
--- a/tests/test_cli_retry.py
+++ b/tests/cli/test_cli_retry.py
@@ -1,6 +1,6 @@
 """Regression tests for CLI /retry history replacement semantics."""
 
-from tests.test_cli_init import _make_cli
+from tests.cli.test_cli_init import _make_cli
 
 
 def test_retry_last_truncates_history_before_requeueing_message():
diff --git a/tests/test_cli_save_config_value.py b/tests/cli/test_cli_save_config_value.py
similarity index 100%
rename from tests/test_cli_save_config_value.py
rename to tests/cli/test_cli_save_config_value.py
diff --git a/tests/test_cli_secret_capture.py b/tests/cli/test_cli_secret_capture.py
similarity index 100%
rename from tests/test_cli_secret_capture.py
rename to tests/cli/test_cli_secret_capture.py
diff --git a/tests/test_cli_skin_integration.py b/tests/cli/test_cli_skin_integration.py
similarity index 100%
rename from tests/test_cli_skin_integration.py
rename to tests/cli/test_cli_skin_integration.py
diff --git a/tests/test_cli_status_bar.py b/tests/cli/test_cli_status_bar.py
similarity index 100%
rename from tests/test_cli_status_bar.py
rename to tests/cli/test_cli_status_bar.py
diff --git a/tests/test_cli_tools_command.py b/tests/cli/test_cli_tools_command.py
similarity index 100%
rename from tests/test_cli_tools_command.py
rename to tests/cli/test_cli_tools_command.py
diff --git a/tests/test_personality_none.py b/tests/cli/test_personality_none.py
similarity index 100%
rename from tests/test_personality_none.py
rename to tests/cli/test_personality_none.py
diff --git a/tests/test_quick_commands.py b/tests/cli/test_quick_commands.py
similarity index 100%
rename from tests/test_quick_commands.py
rename to tests/cli/test_quick_commands.py
diff --git a/tests/test_reasoning_command.py b/tests/cli/test_reasoning_command.py
similarity index 100%
rename from tests/test_reasoning_command.py
rename to tests/cli/test_reasoning_command.py
diff --git a/tests/test_resume_display.py b/tests/cli/test_resume_display.py
similarity index 100%
rename from tests/test_resume_display.py
rename to tests/cli/test_resume_display.py
diff --git a/tests/test_surrogate_sanitization.py b/tests/cli/test_surrogate_sanitization.py
similarity index 100%
rename from tests/test_surrogate_sanitization.py
rename to tests/cli/test_surrogate_sanitization.py
diff --git a/tests/test_worktree.py b/tests/cli/test_worktree.py
similarity index 100%
rename from tests/test_worktree.py
rename to tests/cli/test_worktree.py
diff --git a/tests/test_worktree_security.py b/tests/cli/test_worktree_security.py
similarity index 100%
rename from tests/test_worktree_security.py
rename to tests/cli/test_worktree_security.py
diff --git a/tests/test_codex_execution_paths.py b/tests/cron/test_codex_execution_paths.py
similarity index 100%
rename from tests/test_codex_execution_paths.py
rename to tests/cron/test_codex_execution_paths.py
diff --git a/tests/test_file_permissions.py b/tests/cron/test_file_permissions.py
similarity index 100%
rename from tests/test_file_permissions.py
rename to tests/cron/test_file_permissions.py
diff --git a/tests/test_anthropic_oauth_flow.py b/tests/hermes_cli/test_anthropic_oauth_flow.py
similarity index 100%
rename from tests/test_anthropic_oauth_flow.py
rename to tests/hermes_cli/test_anthropic_oauth_flow.py
diff --git a/tests/test_anthropic_provider_persistence.py b/tests/hermes_cli/test_anthropic_provider_persistence.py
similarity index 100%
rename from tests/test_anthropic_provider_persistence.py
rename to tests/hermes_cli/test_anthropic_provider_persistence.py
diff --git a/tests/test_api_key_providers.py b/tests/hermes_cli/test_api_key_providers.py
similarity index 100%
rename from tests/test_api_key_providers.py
rename to tests/hermes_cli/test_api_key_providers.py
diff --git a/tests/test_atomic_json_write.py b/tests/hermes_cli/test_atomic_json_write.py
similarity index 100%
rename from tests/test_atomic_json_write.py
rename to tests/hermes_cli/test_atomic_json_write.py
diff --git a/tests/test_atomic_yaml_write.py b/tests/hermes_cli/test_atomic_yaml_write.py
similarity index 100%
rename from tests/test_atomic_yaml_write.py
rename to tests/hermes_cli/test_atomic_yaml_write.py
diff --git a/tests/test_auth_codex_provider.py b/tests/hermes_cli/test_auth_codex_provider.py
similarity index 100%
rename from tests/test_auth_codex_provider.py
rename to tests/hermes_cli/test_auth_codex_provider.py
diff --git a/tests/test_auth_commands.py b/tests/hermes_cli/test_auth_commands.py
similarity index 100%
rename from tests/test_auth_commands.py
rename to tests/hermes_cli/test_auth_commands.py
diff --git a/tests/test_auth_nous_provider.py b/tests/hermes_cli/test_auth_nous_provider.py
similarity index 100%
rename from tests/test_auth_nous_provider.py
rename to tests/hermes_cli/test_auth_nous_provider.py
diff --git a/tests/test_codex_models.py b/tests/hermes_cli/test_codex_models.py
similarity index 100%
rename from tests/test_codex_models.py
rename to tests/hermes_cli/test_codex_models.py
diff --git a/tests/test_config_env_expansion.py b/tests/hermes_cli/test_config_env_expansion.py
similarity index 100%
rename from tests/test_config_env_expansion.py
rename to tests/hermes_cli/test_config_env_expansion.py
diff --git a/tests/test_external_credential_detection.py b/tests/hermes_cli/test_external_credential_detection.py
similarity index 100%
rename from tests/test_external_credential_detection.py
rename to tests/hermes_cli/test_external_credential_detection.py
diff --git a/tests/test_gemini_provider.py b/tests/hermes_cli/test_gemini_provider.py
similarity index 100%
rename from tests/test_gemini_provider.py
rename to tests/hermes_cli/test_gemini_provider.py
diff --git a/tests/test_model_normalize.py b/tests/hermes_cli/test_model_normalize.py
similarity index 100%
rename from tests/test_model_normalize.py
rename to tests/hermes_cli/test_model_normalize.py
diff --git a/tests/test_model_provider_persistence.py b/tests/hermes_cli/test_model_provider_persistence.py
similarity index 100%
rename from tests/test_model_provider_persistence.py
rename to tests/hermes_cli/test_model_provider_persistence.py
diff --git a/tests/test_ollama_cloud_auth.py b/tests/hermes_cli/test_ollama_cloud_auth.py
similarity index 100%
rename from tests/test_ollama_cloud_auth.py
rename to tests/hermes_cli/test_ollama_cloud_auth.py
diff --git a/tests/test_plugin_cli_registration.py b/tests/hermes_cli/test_plugin_cli_registration.py
similarity index 100%
rename from tests/test_plugin_cli_registration.py
rename to tests/hermes_cli/test_plugin_cli_registration.py
diff --git a/tests/test_plugins.py b/tests/hermes_cli/test_plugins.py
similarity index 100%
rename from tests/test_plugins.py
rename to tests/hermes_cli/test_plugins.py
diff --git a/tests/test_plugins_cmd.py b/tests/hermes_cli/test_plugins_cmd.py
similarity index 100%
rename from tests/test_plugins_cmd.py
rename to tests/hermes_cli/test_plugins_cmd.py
diff --git a/tests/test_runtime_provider_resolution.py b/tests/hermes_cli/test_runtime_provider_resolution.py
similarity index 100%
rename from tests/test_runtime_provider_resolution.py
rename to tests/hermes_cli/test_runtime_provider_resolution.py
diff --git a/tests/test_setup_model_selection.py b/tests/hermes_cli/test_setup_model_selection.py
similarity index 100%
rename from tests/test_setup_model_selection.py
rename to tests/hermes_cli/test_setup_model_selection.py
diff --git a/tests/run_agent/__init__.py b/tests/run_agent/__init__.py
new file mode 100644
index 0000000000..e69de29bb2
diff --git a/tests/test_1630_context_overflow_loop.py b/tests/run_agent/test_1630_context_overflow_loop.py
similarity index 100%
rename from tests/test_1630_context_overflow_loop.py
rename to tests/run_agent/test_1630_context_overflow_loop.py
diff --git a/tests/test_413_compression.py b/tests/run_agent/test_413_compression.py
similarity index 100%
rename from tests/test_413_compression.py
rename to tests/run_agent/test_413_compression.py
diff --git a/tests/test_860_dedup.py b/tests/run_agent/test_860_dedup.py
similarity index 100%
rename from tests/test_860_dedup.py
rename to tests/run_agent/test_860_dedup.py
diff --git a/tests/test_agent_guardrails.py b/tests/run_agent/test_agent_guardrails.py
similarity index 100%
rename from tests/test_agent_guardrails.py
rename to tests/run_agent/test_agent_guardrails.py
diff --git a/tests/test_agent_loop.py b/tests/run_agent/test_agent_loop.py
similarity index 99%
rename from tests/test_agent_loop.py
rename to tests/run_agent/test_agent_loop.py
index b95ff7808c..bd9e41b91e 100644
--- a/tests/test_agent_loop.py
+++ b/tests/run_agent/test_agent_loop.py
@@ -16,7 +16,7 @@ from unittest.mock import MagicMock
 import pytest
 
 # Ensure repo root is importable
-sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent.parent))
 
 try:
     from environments.agent_loop import (
diff --git a/tests/test_agent_loop_tool_calling.py b/tests/run_agent/test_agent_loop_tool_calling.py
similarity index 99%
rename from tests/test_agent_loop_tool_calling.py
rename to tests/run_agent/test_agent_loop_tool_calling.py
index 74e67c0beb..3b8d6ac598 100644
--- a/tests/test_agent_loop_tool_calling.py
+++ b/tests/run_agent/test_agent_loop_tool_calling.py
@@ -31,7 +31,7 @@ import pytest
 # pytestmark removed — tests skip gracefully via OPENROUTER_API_KEY check on line 59
 
 # Ensure repo root is importable
-_repo_root = Path(__file__).resolve().parent.parent
+_repo_root = Path(__file__).resolve().parent.parent.parent
 if str(_repo_root) not in sys.path:
     sys.path.insert(0, str(_repo_root))
 
diff --git a/tests/test_agent_loop_vllm.py b/tests/run_agent/test_agent_loop_vllm.py
similarity index 99%
rename from tests/test_agent_loop_vllm.py
rename to tests/run_agent/test_agent_loop_vllm.py
index d47478ecbb..d428490941 100644
--- a/tests/test_agent_loop_vllm.py
+++ b/tests/run_agent/test_agent_loop_vllm.py
@@ -30,7 +30,7 @@ import pytest
 import requests
 
 # Ensure repo root is importable
-_repo_root = Path(__file__).resolve().parent.parent
+_repo_root = Path(__file__).resolve().parent.parent.parent
 if str(_repo_root) not in sys.path:
     sys.path.insert(0, str(_repo_root))
 
diff --git a/tests/test_anthropic_error_handling.py b/tests/run_agent/test_anthropic_error_handling.py
similarity index 100%
rename from tests/test_anthropic_error_handling.py
rename to tests/run_agent/test_anthropic_error_handling.py
diff --git a/tests/test_async_httpx_del_neuter.py b/tests/run_agent/test_async_httpx_del_neuter.py
similarity index 100%
rename from tests/test_async_httpx_del_neuter.py
rename to tests/run_agent/test_async_httpx_del_neuter.py
diff --git a/tests/test_compression_boundary.py b/tests/run_agent/test_compression_boundary.py
similarity index 100%
rename from tests/test_compression_boundary.py
rename to tests/run_agent/test_compression_boundary.py
diff --git a/tests/test_compression_persistence.py b/tests/run_agent/test_compression_persistence.py
similarity index 100%
rename from tests/test_compression_persistence.py
rename to tests/run_agent/test_compression_persistence.py
diff --git a/tests/test_compressor_fallback_update.py b/tests/run_agent/test_compressor_fallback_update.py
similarity index 100%
rename from tests/test_compressor_fallback_update.py
rename to tests/run_agent/test_compressor_fallback_update.py
diff --git a/tests/test_context_pressure.py b/tests/run_agent/test_context_pressure.py
similarity index 100%
rename from tests/test_context_pressure.py
rename to tests/run_agent/test_context_pressure.py
diff --git a/tests/test_context_token_tracking.py b/tests/run_agent/test_context_token_tracking.py
similarity index 100%
rename from tests/test_context_token_tracking.py
rename to tests/run_agent/test_context_token_tracking.py
diff --git a/tests/test_dict_tool_call_args.py b/tests/run_agent/test_dict_tool_call_args.py
similarity index 100%
rename from tests/test_dict_tool_call_args.py
rename to tests/run_agent/test_dict_tool_call_args.py
diff --git a/tests/test_exit_cleanup_interrupt.py b/tests/run_agent/test_exit_cleanup_interrupt.py
similarity index 100%
rename from tests/test_exit_cleanup_interrupt.py
rename to tests/run_agent/test_exit_cleanup_interrupt.py
diff --git a/tests/test_fallback_model.py b/tests/run_agent/test_fallback_model.py
similarity index 100%
rename from tests/test_fallback_model.py
rename to tests/run_agent/test_fallback_model.py
diff --git a/tests/test_flush_memories_codex.py b/tests/run_agent/test_flush_memories_codex.py
similarity index 100%
rename from tests/test_flush_memories_codex.py
rename to tests/run_agent/test_flush_memories_codex.py
diff --git a/tests/test_interactive_interrupt.py b/tests/run_agent/test_interactive_interrupt.py
similarity index 98%
rename from tests/test_interactive_interrupt.py
rename to tests/run_agent/test_interactive_interrupt.py
index 8c0d328c24..762621f220 100644
--- a/tests/test_interactive_interrupt.py
+++ b/tests/run_agent/test_interactive_interrupt.py
@@ -23,7 +23,7 @@ logging.basicConfig(level=logging.DEBUG, stream=sys.stderr,
                     format="%(asctime)s [%(threadName)s] %(message)s")
 log = logging.getLogger("interrupt_test")
 
-sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
 
 from unittest.mock import MagicMock, patch
 from run_agent import AIAgent, IterationBudget
diff --git a/tests/test_interrupt_propagation.py b/tests/run_agent/test_interrupt_propagation.py
similarity index 100%
rename from tests/test_interrupt_propagation.py
rename to tests/run_agent/test_interrupt_propagation.py
diff --git a/tests/test_large_tool_result.py b/tests/run_agent/test_large_tool_result.py
similarity index 100%
rename from tests/test_large_tool_result.py
rename to tests/run_agent/test_large_tool_result.py
diff --git a/tests/test_long_context_tier_429.py b/tests/run_agent/test_long_context_tier_429.py
similarity index 100%
rename from tests/test_long_context_tier_429.py
rename to tests/run_agent/test_long_context_tier_429.py
diff --git a/tests/test_openai_client_lifecycle.py b/tests/run_agent/test_openai_client_lifecycle.py
similarity index 100%
rename from tests/test_openai_client_lifecycle.py
rename to tests/run_agent/test_openai_client_lifecycle.py
diff --git a/tests/test_percentage_clamp.py b/tests/run_agent/test_percentage_clamp.py
similarity index 98%
rename from tests/test_percentage_clamp.py
rename to tests/run_agent/test_percentage_clamp.py
index 67d1191494..fcf1e39e54 100644
--- a/tests/test_percentage_clamp.py
+++ b/tests/run_agent/test_percentage_clamp.py
@@ -122,7 +122,7 @@ class TestSourceLinesAreClamped:
     @staticmethod
     def _read_file(rel_path: str) -> str:
         import os
-        base = os.path.dirname(os.path.dirname(__file__))
+        base = os.path.dirname(os.path.dirname(os.path.dirname(__file__)))
         with open(os.path.join(base, rel_path)) as f:
             return f.read()
 
diff --git a/tests/test_primary_runtime_restore.py b/tests/run_agent/test_primary_runtime_restore.py
similarity index 100%
rename from tests/test_primary_runtime_restore.py
rename to tests/run_agent/test_primary_runtime_restore.py
diff --git a/tests/test_provider_fallback.py b/tests/run_agent/test_provider_fallback.py
similarity index 100%
rename from tests/test_provider_fallback.py
rename to tests/run_agent/test_provider_fallback.py
diff --git a/tests/test_provider_parity.py b/tests/run_agent/test_provider_parity.py
similarity index 100%
rename from tests/test_provider_parity.py
rename to tests/run_agent/test_provider_parity.py
diff --git a/tests/test_real_interrupt_subagent.py b/tests/run_agent/test_real_interrupt_subagent.py
similarity index 100%
rename from tests/test_real_interrupt_subagent.py
rename to tests/run_agent/test_real_interrupt_subagent.py
diff --git a/tests/test_redirect_stdout_issue.py b/tests/run_agent/test_redirect_stdout_issue.py
similarity index 100%
rename from tests/test_redirect_stdout_issue.py
rename to tests/run_agent/test_redirect_stdout_issue.py
diff --git a/tests/test_run_agent.py b/tests/run_agent/test_run_agent.py
similarity index 100%
rename from tests/test_run_agent.py
rename to tests/run_agent/test_run_agent.py
diff --git a/tests/test_run_agent_codex_responses.py b/tests/run_agent/test_run_agent_codex_responses.py
similarity index 100%
rename from tests/test_run_agent_codex_responses.py
rename to tests/run_agent/test_run_agent_codex_responses.py
diff --git a/tests/test_session_meta_filtering.py b/tests/run_agent/test_session_meta_filtering.py
similarity index 100%
rename from tests/test_session_meta_filtering.py
rename to tests/run_agent/test_session_meta_filtering.py
diff --git a/tests/test_session_reset_fix.py b/tests/run_agent/test_session_reset_fix.py
similarity index 98%
rename from tests/test_session_reset_fix.py
rename to tests/run_agent/test_session_reset_fix.py
index ee65ed90d1..1fd1223ced 100644
--- a/tests/test_session_reset_fix.py
+++ b/tests/run_agent/test_session_reset_fix.py
@@ -13,7 +13,7 @@ from pathlib import Path
 import pytest
 
 # Ensure repo root is importable
-sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent.parent))
 
 # Stub out optional heavy dependencies not installed in the test environment
 sys.modules.setdefault("fire", types.SimpleNamespace(Fire=lambda *a, **k: None))
diff --git a/tests/test_streaming.py b/tests/run_agent/test_streaming.py
similarity index 100%
rename from tests/test_streaming.py
rename to tests/run_agent/test_streaming.py
diff --git a/tests/test_strict_api_validation.py b/tests/run_agent/test_strict_api_validation.py
similarity index 100%
rename from tests/test_strict_api_validation.py
rename to tests/run_agent/test_strict_api_validation.py
diff --git a/tests/test_token_persistence_non_cli.py b/tests/run_agent/test_token_persistence_non_cli.py
similarity index 100%
rename from tests/test_token_persistence_non_cli.py
rename to tests/run_agent/test_token_persistence_non_cli.py
diff --git a/tests/test_tool_arg_coercion.py b/tests/run_agent/test_tool_arg_coercion.py
similarity index 100%
rename from tests/test_tool_arg_coercion.py
rename to tests/run_agent/test_tool_arg_coercion.py
diff --git a/tests/test_managed_server_tool_support.py b/tests/tools/test_managed_server_tool_support.py
similarity index 100%
rename from tests/test_managed_server_tool_support.py
rename to tests/tools/test_managed_server_tool_support.py
diff --git a/tests/test_tool_call_parsers.py b/tests/tools/test_tool_call_parsers.py
similarity index 100%
rename from tests/test_tool_call_parsers.py
rename to tests/tools/test_tool_call_parsers.py