Testing

The test suite is a single end-to-end suite under tests/e2e/. There is one test directory, no unit-only tests, and no importorskip / skipif guards. The full HiveMind 2.x stack is a hard [e2e] dependency.

Running the tests

uv pip install -e .[e2e]
pytest tests/

What the e2e suite exercises

Tests boot a real hivemind-core master in-process, through the hivescope loopback hub, and drive the satellite's real code over a real HiveMessageBusClient across a localhost WebSocket. Only the hardware seams are mocked.

test_satellite_hivemind_e2e.py: real satellite over a real hub

Test	Path exercised
test_microphone_stream_reaches_hub_binary	Real HiveMindMicrophoneClient.run() capture loop, fed by a fake mic and a fake VAD, streams HiveMessageType.BINARY / RAW_AUDIO frames to the real hub
test_client_constructs_with_mocked_hardware	Real client constructs, handshakes with the hub, and binds handlers, with mic/VAD injected and media/PHAL disabled
test_hub_speak_triggers_tts_request	Hub agent speak routed to the peer, through real handle_speak, sends a speak:synth request back to the hub
test_b64_tts_audio_queued_for_playback	speak:b64_audio.response reaches real handle_speak_b64, which decodes the audio and enqueues it on the PlaybackThread queue

test_bridge1_conformance.py: OVOS-BRIDGE-1 / SESSION-1 conformance

This test asserts envelope, source stamping, destination routing, session fidelity, and FIFO ordering against the real hub.

test_acl.py: ACL policy admission

This test checks allowed_types denial, skill-blacklist injection, and reserved-session rejection through the hub's policy chain.

How the hardware is mocked

The satellite touches three hardware seams. Each is replaced in-process with no device, model, or network beyond localhost.

• Microphone: a patch replaces OVOSMicrophoneFactory.create with a FakeMicrophone that yields a fixed run of non-empty chunks, then silence. No ALSA / PortAudio device is opened.
• VAD: a patch replaces OVOSVADFactory.create with a FakeVAD that reports speech for the first N reads, then silence, driving the speech-start → stream → silence-timeout branch of run().
• Audio playback: the client is constructed with enable_media=False, so no AudioService (and no sound device) is created, and PHAL is absent. The PlaybackThread never starts; the tests inspect queued playback items directly instead of reaching a speaker.

Everything else is genuine production code: the VAD-gated capture loop, encryption, the HiveMessageBusClient, the WebSocket transport, hivemind-core, the binary protocol, the agent bus, and the policy chain.

CI

The e2e suite runs on every PR through .github/workflows/e2e_tests.yml, which calls the shared OpenVoiceOS/gh-automations build-tests workflow with install_extras: e2e and test_path: tests/e2e/. build_tests.yml also clean-installs the package across Python 3.10-3.13. coverage.yml reports coverage of hivemind_mic_sat.

---

← Deployment · Home · Troubleshooting →