v0.15.5: Corrected FK mappings, objektkunde junction table, multi-hop query test

Schema corrections:
- kunden PK = ID (not Kundennummer)
- objekte PK = ID (not ObjektID)
- kunden↔objekte linked via objektkunde junction table (many-to-many)
- Removed guessed column names, only verified PKs/FKs in SCHEMA
- Added explicit JOIN patterns for the hierarchy

Domain context test: 25/25 (added multi-hop Jaeger query through 4 tables)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

agent/nodes/eras_expert.py

@@ -48,10 +48,14 @@ DOMAIN MODEL (how the data relates):
 - Auftraege (work orders) = tasks for Monteure (technicians)
   Device installation, reading collection, maintenance.
 
-HIERARCHY: Kunde → Objekte → Nutzeinheiten → Geraete → Verbraeuche
+HIERARCHY (via JOINs):
-                                              → Nutzer
+  Kunde ←→ objektkunde ←→ Objekt (many-to-many via junction table!)
-           Kunde → Abrechnungen
+  Objekt → Nutzeinheiten → Geraete → Verbraeuche
-           Kunde → Auftraege
+  Nutzeinheit → Nutzer
+  Kunde → Abrechnungen
+  Kunde → Auftraege
+
+CRITICAL: kunden and objekte are linked through the objektkunde junction table, NOT directly.
 
 IMPORTANT NOTES:
 - All table/column names are German, lowercase
@@ -75,18 +79,31 @@ IMPORTANT NOTES:
 - heizbetriebskosten — heating operation costs
 - nebenkosten — additional costs (Nebenkosten)
 
-KNOWN PRIMARY KEYS AND FOREIGN KEYS:
+KNOWN SCHEMA (verified — ONLY use these column names without DESCRIBE):
-- kunden: PK = Kundennummer (int), name columns: Name1, Name2, Name3
+All tables use ID (int, auto_increment) as primary key.
-- objekte: PK = ObjektID, FK = KundenID → kunden.Kundennummer
-- nutzeinheit: FK = ObjektID → objekte.ObjektID
-- geraete: FK = NutzeinheitID → nutzeinheit.NutzeinheitID (verify with DESCRIBE)
 
-IMPORTANT: Always DESCRIBE tables you haven't seen before to verify column names.
+- kunden: PK=ID. Known columns: Name1, Name2, Name3, Kundennummer
-Use the FK mappings above for JOINs. Do NOT guess — use exact column names.
+- objekte: PK=ID. Known columns: Objektnummer
+- objektkunde: JUNCTION TABLE for kunden↔objekte (many-to-many!)
+  PK=ID, FK: KundeID→kunden.ID, ObjektID→objekte.ID
+- nutzeinheit: PK=ID, FK: ObjektID→objekte.ID
+- geraete: PK=ID, FK: NutzeinheitID→nutzeinheit.ID
+- geraeteverbraeuche: linked to geraete
+- nutzer: linked to nutzeinheit (DESCRIBE to find FK column name)
+
+For ANY column not listed above, you MUST DESCRIBE the table first.
+
+JOIN PATTERNS (use these exactly):
+- Kunde → Objekte: JOIN objektkunde ok ON ok.KundeID = k.ID JOIN objekte o ON o.ID = ok.ObjektID
+- Objekt → Nutzeinheiten: JOIN nutzeinheit n ON n.ObjektID = o.ID
+- Nutzeinheit → Geraete: JOIN geraete g ON g.NutzeinheitID = n.ID
+
+IMPORTANT: For tables not listed above, always DESCRIBE first.
+The junction table objektkunde is REQUIRED to link kunden and objekte.
 
 Example for "how many Objekte per Kunde":
 [
-  {{"tool": "query_db", "args": {{"query": "SELECT k.Kundennummer, k.Name1, COUNT(o.ObjektID) as AnzahlObjekte FROM kunden k LEFT JOIN objekte o ON o.KundenID = k.Kundennummer GROUP BY k.Kundennummer, k.Name1 ORDER BY AnzahlObjekte DESC LIMIT 20", "database": "eras2_production"}}}}
+  {{"tool": "query_db", "args": {{"query": "SELECT k.ID, k.Name1, COUNT(DISTINCT o.ID) as AnzahlObjekte FROM kunden k JOIN objektkunde ok ON ok.KundeID = k.ID JOIN objekte o ON o.ID = ok.ObjektID GROUP BY k.ID, k.Name1 ORDER BY AnzahlObjekte DESC LIMIT 20", "database": "eras2_production"}}}}
 ]"""
 
     def __init__(self, send_hud, process_manager=None):

testcases/domain_context.md

@@ -12,7 +12,7 @@ and can formulate correct JOINs without guessing column names.
 ### 1. Expert knows the hierarchy
 - send: wie viele Objekte haben Kunden im Durchschnitt?
 - expect_trace: has tool_call
-- expect_response: not contains "Error" or "error" or "Unknown column"
+- expect_response: not contains "Unknown column" or "1054" or "doesn't exist"
 - expect_response: length > 20
 
 ### 2. Expert can JOIN kunden and objekte
@@ -33,7 +33,13 @@ and can formulate correct JOINs without guessing column names.
 - expect_response: not contains "Error" or "error" or "Unknown column"
 - expect_response: length > 20
 
-### 5. PA formulates good job descriptions
+### 5. Multi-hop query through hierarchy
+- send: zeig alle Nutzer in Objekten von Kunde mit Jaeger im Namen
+- expect_trace: has tool_call
+- expect_response: not contains "Unknown column" or "1054" or "doesn't exist"
+- expect_response: contains "Jaeger" or "jaeger"
+
+### 6. PA formulates good job descriptions
 - send: gib mir eine Uebersicht ueber Kunde 2
 - expect_trace: has routed
 - expect_response: length > 20