Vorher: _execute_tool (agent_loop.py) und der KI-Chat-Loop
(stream_chat_comm) führten JEDES im Registry registrierte Tool aus, wenn
das LLM dessen Namen lieferte — ohne Abgleich mit der angebotenen Liste,
ohne required_permission-Check. Reproduktion (Astra): Nur audit_allowed
angeboten, Modell nannte audit_restricted (system:admin) → Handler lief.
Fix (fail-closed, an ALLEN Ausfuehrungspfaden):
- _check_tool_access: (1) Allowlist — nur Tools die dem LLM angeboten
wurden duerfen laufen; (2) required_permission gegen die AKTUELLEN
User-Rechte (deny-first, Rechteentzug wirkt sofort, ohne Kontext =
Ablehnung). Guard vor dry-run/approval/execute-Pfaden.
- stream_chat_comm: gleicher Allowlist-Guard vor execute_tool_call.
- run_react_loop/agent_runner/agent_stream/agent_routes reichen
user_permissions durch (perm_ctx bzw. Session-User).
- check_permission: Session-Kontexte tragen denied_permissions statt
denied — beide Keys werden gelesen, Deny-Liste wird nie mehr ignoriert.
Tests: test_agent_loop.py 18/18 (7 neue F01-Tests nach Astra-Abnahme:
nicht angeboten → Handler null; fehlende Permission → abgewiesen;
Fail-closed ohne Kontext; Deny-Liste session-shape; Rechteentzug
mitten im Lauf wirkt auf naechste Aktion; dry-run guardet auch).
ruff clean. Pre-existing-Beweis: permission_system_live-Failures
reproduzieren sich ohne diesen Patch identisch (Plugin-Aktivierung in
ephemeraler Test-DB, bekanntes Vorbestands-Finding).
Migration 0137: Drop AI chat tables (ai_chat_sessions, ai_chat_messages, ai_chat_attachments, ai_conversations, ai_messages)
Backend:
- Remove AIChatSession, AIChatMessage, AIChatAttachment models from ai_assistant/models.py
- Remove AIConversation, AIMessage from app/models/__init__.py
- Remove session/message/stream/attachment routes from ai_assistant/routes.py
- Add new streaming route POST /ai/conversations/{conversation_id}/stream using comm tables
- Add new messages route GET /ai/conversations/{conversation_id}/messages using comm tables
- Add stream_chat_comm, get_comm_messages, save_comm_message to services.py
- Update external_api.py to use CommConversation/CommMessage instead of AIChatSession/AIChatMessage
- Update unified_search ai_chat_provider to search comm_messages with conversation_type=ai
- Remove ai_copilot router from main.py and routes/__init__.py
- Remove ai_conversation from entity_permissions.py and owner_transfer_service.py
- Update ai_assistant/plugin.py get_entity_models to remove AIChatSession
- Guard ai_copilot_service.py imports with try/except
Frontend:
- Remove AIAssistant.tsx, AIAssistantStandalone.tsx, SessionList.tsx, ChatWindow.tsx
- Remove AI Assistant routes from routes/index.tsx
- Update api/ai.ts: streamChat uses /ai/conversations/{id}/stream, fetchMessages uses /ai/conversations/{id}/messages
- Update Communication.tsx: use convId for AI streaming, remove aiSessionId, use fetchAiMessages for AI conversations
- Update AiChatPanel.tsx: create comm conversation instead of AI session, use new fetchMessages
- Update AISidebar.tsx: remove ChatWindow import, show placeholder
tsc clean, build successful, backend import OK
- Create call_crm_api tool: single tool that can call ANY CRM API endpoint
- Inject OpenAPI spec into system prompt so AI knows all available endpoints
- Always include call_crm_api in agent tools (not just via tool_ids)
- Extend get_current_user to support internal header-based auth (X-Internal-Call,
X-Tenant-Id, X-User-Id) for AI tool API access
- No more manual tool-per-endpoint registration needed
Agent Zero