Testing

WebChat is a Python + JavaScript hybrid, so it has two test suites, one for each half. There is a single Python test directory (tests/, with the e2e suite under tests/e2e/). There are no importorskip / skipif guards. The full HiveMind 2.x stack is a hard [e2e] dependency.

Python tests

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

• tests/test_smoke.py: import, version, Tornado server construction without binding a socket, and bridge construction with a mocked client (no connect).
• tests/e2e/: the real end-to-end suite below.

Python e2e: real backend over a real hub

tests/e2e/test_webchat_hivemind_e2e.py boots a real hivemind-core master in-process (via the hivescope loopback hub) and drives the webchat's real WebchatBridge over a real HiveMessageBusClient across a localhost WebSocket. Only the browser / websocket frontend is mocked. The browser does not run, so user input is injected by calling the bridge's say() (exactly what app.js does on the wire) and rendered speak replies are captured from handle_speak instead of being painted into the DOM.

Test	Path exercised
test_chat_message_reaches_hub_as_utterance	(mocked browser) text → real WebchatBridge.say() → real hub decrypts + injects recognizer_loop:utterance on the agent bus
test_bridge_constructs_with_real_handshake	Real bridge builds a real HiveMessageBusClient, handshakes with the hub, binds its speak handler
test_speak_response_routed_back_to_webchat	Hub agent speak routed to the peer → real handle_speak → rendered back to the (mocked) browser chat log
test_full_chat_roundtrip	Full loop: message browser → bridge → hub, the hub answers with a speak routed back → bridge → browser

Everything between the backend and the hub (the bridge, encryption, the HiveMessageBusClient, the WebSocket transport, the hivemind-core listener, agent bus, and reverse routing) is genuine production code.

JavaScript e2e: real browser client over a real hub

npm install   # ws (the only dev-dependency)
npm test      # node tests/e2e.mjs

tests/e2e.mjs loads the exact HiveMind-js V1 client the page ships (sibling checkout if present, else fetched from the same CDN URL used in production), polyfills globalThis.WebSocket with ws, spawns a real loopback hivemind-core hub (tests/hub_fixture.py via hivescope), runs the full V1 handshake (PBKDF2 + AES-GCM), sends an encrypted utterance, and asserts the hub decrypts and injects it. This needs a Python interpreter with the [e2e] stack on PATH (PYTHON=... overrides it).

This is the browser-side round-trip the Python suite intentionally mocks: the two suites together cover both halves of the hybrid.

CI

Workflow	Covers
build_tests.yml	Build + clean-install across Python 3.10–3.13 (install_extras: e2e), plus the smoke tests
e2e_tests.yml	The Python e2e suite (tests/e2e/) against a real loopback hub, single Python
js_e2e_tests.yml	The Node JS e2e (npm test) against a real loopback hub
coverage.yml	Coverage of hivemind_webchat
lint.yml	Ruff
license_tests.yml, pip_audit.yml, release_preview.yml, repo_health.yml	License / vuln / release / health checks

All Python CI uses the shared OpenVoiceOS/gh-automations reusable workflows @dev.

---

← Dependencies · Home · Troubleshooting →