Try splitting tests that don't need hosted runner

.github/workflows/ci.yml

@@ -14,7 +14,68 @@ permissions:
   packages: read
 
 jobs:
-  ci-complete:
+  tests:
+    timeout-minutes: 15
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v6
+      - name: Install uv
+        uses: astral-sh/setup-uv@v6
+        with:
+          enable-cache: true
+      - name: Setup Python
+        uses: actions/setup-python@v6
+        with:
+          python-version: 3.12
+
+      - name: Install system dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y libturbojpeg0-dev portaudio19-dev build-essential \
+            libgl1 libgl1-mesa-dri python3-dev cyclonedds-dev
+          sudo rm -f /usr/bin/git-lfs
+
+      - name: Configure LCM networking
+        run: |
+          sudo ip link set lo multicast on
+          sudo ip route add 224.0.0.0/4 dev lo || true
+          sudo sysctl -w net.core.rmem_max=67108864
+          sudo sysctl -w net.core.rmem_default=67108864
+
+      - name: Install Python dependencies
+        run: uv sync --only-group tests --frozen
+
+      - name: Build C++ extensions
+        run: |
+          source .venv/bin/activate
+          uv pip install pybind11
+          python setup.py build_ext --inplace
+
+      - name: Remove pydrake stubs
+        run: |
+          find .venv/lib/*/site-packages/pydrake -name '*.pyi' -delete 2>/dev/null || true
+
+      - name: Run tests
+        run: |
+          source .venv/bin/activate
+          _DIMOS_COV=1 coverage run -m pytest --durations=0 && coverage html && coverage report
+
+      - name: Run mypy
+        if: ${{ !cancelled() }}
+        run: |
+          source .venv/bin/activate
+          mypy dimos
+
+      - name: Upload coverage report
+        if: github.event_name == 'push' && !cancelled()
+        uses: actions/upload-artifact@v5
+        with:
+          name: coverage-report
+          path: htmlcov/
+
+  slow-tests:
     if: github.event_name != 'pull_request' || github.event.pull_request.draft == false
     timeout-minutes: 60
     runs-on: [self-hosted, Linux]
@@ -36,35 +97,34 @@ jobs:
           git clean -ffdx -e .venv
 
       - name: Install Python dependencies
-        run: uv sync --all-extras --no-extra dds --no-extra unitree-dds --frozen
+        run: uv sync --only-group tests-self-hosted --frozen
 
       - name: Remove pydrake stubs
         run: |
           find .venv/lib/*/site-packages/pydrake -name '*.pyi' -delete 2>/dev/null || true
 
       - name: Run tests
-        if: github.event_name != 'push'
-        run: |
-          /entrypoint.sh bash -c "source .venv/bin/activate && pytest --durations=0 -m 'not (tool or mujoco)'"
-
-      - name: Run tests with coverage
-        if: github.event_name == 'push'
         run: |
-          /entrypoint.sh bash -c "source .venv/bin/activate && _DIMOS_COV=1 coverage run -m pytest --durations=0 -m 'not (tool or mujoco)' && coverage combine && coverage html && coverage report"
+          /entrypoint.sh bash -c "source .venv/bin/activate && pytest --durations=0 -m '(slow or skipif_no_ros) and not (tool or mujoco)'"
 
       - name: Run mypy
         if: ${{ !cancelled() }}
         run: |
           /entrypoint.sh bash -c "source .venv/bin/activate && MYPYPATH=/opt/ros/humble/lib/python3.10/site-packages mypy dimos"
 
-      - name: Upload coverage report
-        if: github.event_name == 'push' && !cancelled()
-        uses: actions/upload-artifact@v5
-        with:
-          name: coverage-report
-          path: htmlcov/
-
       - name: Check disk space
         if: failure()
         run: |
           df -h
+
+  ci-gate:
+    if: always()
+    needs: [tests, slow-tests]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check results
+        run: |
+          if [[ "${{ needs.tests.result }}" != "success" ]] || \
+             [[ "${{ needs.slow-tests.result }}" != "success" ]]; then
+            exit 1
+          fi

.github/workflows/macos.yml

@@ -69,7 +69,7 @@ jobs:
 
       - name: Install dependencies
         run: |
-          uv sync --all-extras --no-extra dds --no-extra unitree-dds --no-extra cuda --frozen
+          uv sync --only-group tests-self-hosted --frozen
 
       - name: Build C++ extensions
         run: |

dimos/agents/mcp/test_mcp_client.py

@@ -30,7 +30,6 @@ def add(self, x: int, y: int) -> str:
         return str(x + y)
 
 
-@pytest.mark.slow
 def test_can_call_tool(agent_setup):
     history = agent_setup(
         blueprints=[Adder.blueprint()],
@@ -63,7 +62,6 @@ def register_user(self, name: str) -> str:
         return "User name registered successfully."
 
 
-@pytest.mark.slow
 def test_can_call_again_on_error(agent_setup):
     history = agent_setup(
         blueprints=[UserRegistration.blueprint()],
@@ -113,7 +111,6 @@ def go_to_location(self, description: str) -> str:
         return f"Going to the {description}."
 
 
-@pytest.mark.slow
 def test_multiple_tool_calls_with_multiple_messages(agent_setup):
     history = agent_setup(
         blueprints=[MultipleTools.blueprint(), NavigationSkill.blueprint()],
@@ -167,7 +164,6 @@ def test_multiple_tool_calls_with_multiple_messages(agent_setup):
     assert len(go_to_location_calls) == 2
 
 
-@pytest.mark.slow
 def test_prompt(agent_setup):
     history = agent_setup(
         blueprints=[],

dimos/agents/skills/test_google_maps_skill_container.py

@@ -16,7 +16,6 @@
 from typing import Any
 
 from langchain_core.messages import HumanMessage
-import pytest
 
 from dimos.agents.skills.google_maps_skill_container import GoogleMapsSkillContainer
 from dimos.core.module import Module
@@ -71,7 +70,6 @@ def __init__(self, **kwargs: Any):
         self._max_valid_distance = 20000
 
 
-@pytest.mark.slow
 def test_where_am_i(agent_setup) -> None:
     history = agent_setup(
         blueprints=[FakeGPS.blueprint(), MockedWhereAmISkill.blueprint()],
@@ -81,7 +79,6 @@ def test_where_am_i(agent_setup) -> None:
     assert "bourbon" in history[-1].content.lower()
 
 
-@pytest.mark.slow
 def test_get_gps_position_for_queries(agent_setup) -> None:
     history = agent_setup(
         blueprints=[FakeGPS.blueprint(), MockedPositionSkill.blueprint()],

dimos/agents/skills/test_gps_nav_skills.py

@@ -13,7 +13,6 @@
 # limitations under the License.
 
 from langchain_core.messages import HumanMessage
-import pytest
 
 from dimos.agents.skills.gps_nav_skill import GpsNavSkillContainer
 from dimos.core.core import rpc
@@ -40,7 +39,6 @@ class MockedGpsNavSkill(GpsNavSkillContainer):
     _max_valid_distance = 50000
 
 
-@pytest.mark.slow
 def test_set_gps_travel_points(agent_setup) -> None:
     history = agent_setup(
         blueprints=[
@@ -59,7 +57,6 @@ def test_set_gps_travel_points(agent_setup) -> None:
     assert "success" in history[-1].content.lower()
 
 
-@pytest.mark.slow
 def test_set_gps_travel_points_multiple(agent_setup) -> None:
     history = agent_setup(
         blueprints=[

dimos/agents/skills/test_navigation.py

@@ -15,7 +15,6 @@
 from typing import Any
 
 from langchain_core.messages import HumanMessage
-import pytest
 
 from dimos.agents.skills.navigation import NavigationSkillContainer
 from dimos.core.core import rpc
@@ -118,7 +117,6 @@ def _navigate_using_semantic_map(self, query):
         return f"Successfuly arrived at '{query}'"
 
 
-@pytest.mark.slow
 def test_stop_movement(agent_setup) -> None:
     history = agent_setup(
         blueprints=[
@@ -133,7 +131,6 @@ def test_stop_movement(agent_setup) -> None:
     assert "stopped" in history[-1].content.lower()
 
 
-@pytest.mark.slow
 def test_start_exploration(agent_setup) -> None:
     history = agent_setup(
         blueprints=[
@@ -150,7 +147,6 @@ def test_start_exploration(agent_setup) -> None:
     assert "explor" in history[-1].content.lower()
 
 
-@pytest.mark.slow
 def test_go_to_semantic_location(agent_setup) -> None:
     history = agent_setup(
         blueprints=[

dimos/agents/skills/test_unitree_skill_container.py

@@ -16,7 +16,6 @@
 from typing import Any
 
 from langchain_core.messages import HumanMessage
-import pytest
 
 from dimos.core.core import rpc
 from dimos.core.module import Module
@@ -53,7 +52,6 @@ class MockedUnitreeSkill(UnitreeSkillContainer):
     pass
 
 
-@pytest.mark.slow
 def test_pounce(agent_setup) -> None:
     history = agent_setup(
         blueprints=[