Expand test_run_service.py from a single stub to a full test suite
using an in-memory SQLite database (same pattern as test_blueprint_service.py).
New test classes cover:
- TestCreateRun: verifies field storage, blueprint_id, execution_mode, created_at
- TestGetRun: existing run retrieval and None for unknown ID
- TestListRuns: empty list, full listing, limit and offset pagination
- TestUpdateRun: status changes, auto-setting completed_at on terminal
statuses (completed/failed), not setting it for non-terminal statuses,
storing final_draft/critic_score/iteration_count, and silently ignoring
unknown field keys via the hasattr guard
All tests use isolated fixtures that create and drop tables per test,
ensuring no state leaks between cases.
https://claude.ai/code/session_01MLq4Vjq9XWKPtu3iX3iqL3
- Replace useState+useEffect sync pattern with useMemo in
EdgeSettingsPanel.tsx to eliminate cascading re-renders
- Remove redundant draft state in NodeSettingsPanel.tsx and use
store data directly, eliminating useEffect sync loop
- Fix mock paths in test_tools.py: patch tavily.TavilyClient and
pypdf.PdfReader at their source modules (lazy imports)
- Remove unused variable assignment in routes.py (god mode reject)
- Remove unused node_lookup dicts in dynamic_graph_builder.py
- Remove unused imports across test files (Blueprint, CouncilState,
pytest, llm assignments)
- Remove unused CouncilBlueprint type import in types.test.ts
- Run npm audit fix to resolve moderate vulnerability
All 107 backend tests and 26 frontend tests pass. Ruff, ESLint,
and TypeScript checks are clean.
https://claude.ai/code/session_01XqzyT6fhS8sUe9P5fCmuVU
Backend:
- Add Tavily web search tool wrapper (tools/web_search.py)
- Add PDF reader + ChromaDB vector store tool (tools/pdf_reader.py)
- Bind tools to LLM calls via .bind_tools() in dynamic_graph_builder
- Implement God Mode using LangGraph interrupt_before + MemorySaver
- Add approve/reject/modify API endpoints for God Mode
- Add PDF upload endpoint with ingestion pipeline
- Add persistent run history (CouncilRun model + run_service + API)
- Add Alembic migration for council_runs table
- Enhance WebSocket to emit run_paused and run_resumed events
- Add tests for tools, God Mode, and run history
Frontend:
- Add God Mode approval UI (GodModePanel component)
- Add Auto-Pilot / God Mode toggle in Konferenzzimmer
- Add functional PDF upload handler
- Add Conditional Edge editor (EdgeSettingsPanel component)
- Add edge click selection in ArchitectCanvas
- Update Zustand store with edge selection and update actions
- Update types for God Mode, execution modes, and WS events
- Update API client with God Mode, PDF upload, and blueprint run endpoints
- Update WebSocket hook for paused/resumed events
- Add Vitest config and frontend tests (store, parser, types, API)
https://claude.ai/code/session_017U6idFgaqnYTXzPxA7mxMv
- Add dynamic_graph_builder.py that constructs LangGraph graphs at runtime
from frontend CouncilBlueprint JSON (no more hardcoded graphs in production)
- Add PostgreSQL persistence via SQLAlchemy async with Blueprint model
- Add blueprint CRUD endpoints (POST/GET/PUT/DELETE /api/councils/)
- Add POST /api/councils/{id}/run to execute blueprints dynamically
- Add Alembic migration infrastructure with initial blueprints table
- Add database.py with async engine and SQLite fallback for dev/test
- Fix missing typing-extensions and add aiosqlite dependency
- Add 42 new tests (80/80 total passing) covering dynamic graph building,
blueprint service CRUD, and API integration
https://claude.ai/code/session_014yZUxrPsgZbvkebXbCXR4U
