erp-nutzfahrzeuge/test_report.md

Test Report — T07: KI-Copilot

Backend Tests

Command: cd backend && python -m pytest tests/test_copilot.py tests/test_copilot_coverage.py --cov=app.services.copilot_service --cov=app.routers.copilot --cov-report=term-missing -v

Result: 48 passed, 0 failed

Coverage

Module	Stmts	Miss	Cover
app/routers/copilot.py	33	8	76%
app/services/copilot_service.py	149	10	93%
TOTAL	182	18	90%

Coverage exceeds the 80% requirement.

Test Categories

• TestCopilotChat (7 tests): POST /copilot/chat — valid message, empty message 422, no actions, DB persistence, contact search action, auth required, session continuation
• TestCopilotAction (5 tests): POST /copilot/action — search_vehicles, search_contacts, invalid action 400, get_sale_overview, auth required
• TestCopilotHistory (4 tests): GET /copilot/history — pagination, empty, auth required, session filter
• TestCopilotVoice (3 tests): POST /copilot/voice — transcription+response, empty audio 422, auth required
• TestCopilotServiceUnit (9 tests): _parse_ai_response, system prompt, action registry, execute_action, _get_or_create_session
• TestCopilotServiceCoverage (20 tests): Direct service-level tests for chat(), get_history(), get_sessions(), execute_confirmed_action(), transcribe_audio(), voice_chat(), _call_openrouter_chat()

Frontend Tests

Command: cd frontend && npx vitest run tests/copilot.test.tsx

Result: 24 passed, 0 failed

Test Categories

• ChatInput (5 tests): renders, sends, clears, empty prevention, disabled state
• MessageList (4 tests): placeholder, user/assistant messages, loading indicator, action proposals
• ActionPreview (5 tests): renders, confirm, dismiss, null when empty, action labels
• ChatHistory (4 tests): title, empty state, session grouping, session selection
• ChatInterface (6 tests): renders, sends+displays, action preview, action execution, error display, history toggle

Smoke Test

• Backend import verified: from app.main import app succeeds
• Copilot router registered in main.py under /api/v1/copilot
• All endpoints accessible: POST /chat, POST /action, GET /history, POST /voice
• OpenRouter mocked in all tests — no real API calls
• System prompt contains ERP context (Fahrzeugtypen, Kontakttypen, Verkaufsworkflow) and all 5 actions
• Action system: Copilot proposes actions, user confirms via /action endpoint
• Chat history persisted to DB (CopilotSession + CopilotChat models)
• Voice endpoint accepts base64 audio, transcribes, returns chat response