Harden QuoteJudge managed persistence

a2a.yaml

@@ -1,5 +1,5 @@
 name: quote-judge-studio-v1
-version: 0.1.0
+version: 0.1.4
 entrypoint: agent:QuoteJudgeStudioV1
 expose:
   public: false

agent.py

@@ -54,10 +54,10 @@ ALLOWED_UPLOAD_MEDIA_TYPES = {
     "application/vnd.ms-excel",
     "text/plain",
 }
-
-# Local/dev fallback only. Hosted production uses DATABASE_URL and managed Postgres.
-_LOCAL_COMPARISONS: dict[tuple[str, str], dict[str, Any]] = {}
-_LOCAL_RECEIPTS: list[dict[str, Any]] = []
+AGENT_VERSION = "0.1.4"
+DB_CONNECT_TIMEOUT_SECONDS = 5
+DB_STATEMENT_TIMEOUT_MS = 10_000
+DB_LOCK_TIMEOUT_MS = 5_000
 
 
 class QuoteJudgeStudioV1Config(BaseModel):
@@ -128,7 +128,7 @@ class QuoteJudgeStudioV1(A2AAgent[QuoteJudgeStudioV1Config, PlatformUserAuth]):
         "scoring, persists per-user comparisons, supports browser uploads, and serves "
         "a polished packed product UI."
     )
-    version = "0.1.3"
+    version = AGENT_VERSION
 
     config_model = QuoteJudgeStudioV1Config
     auth_model = PlatformUserAuth
@@ -229,6 +229,12 @@ class QuoteJudgeStudioV1(A2AAgent[QuoteJudgeStudioV1Config, PlatformUserAuth]):
         record = await asyncio.to_thread(_load_comparison, tenant, comparison_id)
         if record is None:
             return {"ok": False, "code": "comparison_not_found", "comparison_id": comparison_id}
+        if record.get("ok") is False:
+            return {
+                "ok": False,
+                "code": record.get("code", "persistence_unavailable"),
+                "comparison_id": comparison_id,
+            }
         result = dict(record["result"])
         result["comparison_id"] = comparison_id
         result["ok"] = True
@@ -477,7 +483,15 @@ def _connect() -> Any:
     url = _database_url()
     if not url:
         raise RuntimeError("DATABASE_URL is not configured")
-    return psycopg.connect(url, row_factory=dict_row)
+    return psycopg.connect(
+        url,
+        row_factory=dict_row,
+        connect_timeout=DB_CONNECT_TIMEOUT_SECONDS,
+        options=(
+            f"-c statement_timeout={DB_STATEMENT_TIMEOUT_MS} "
+            f"-c lock_timeout={DB_LOCK_TIMEOUT_MS}"
+        ),
+    )
 
 
 def _persist_comparison(
@@ -487,13 +501,7 @@ def _persist_comparison(
     receipt: dict[str, Any],
 ) -> dict[str, Any]:
     if not _database_url() or psycopg is None:
-        _LOCAL_COMPARISONS[(tenant_key, comparison_id)] = {
-            "result": json.loads(json.dumps(result)),
-            "updated_at": receipt["created_at"],
-            "latest_receipt_id": receipt["receipt_id"],
-        }
-        _LOCAL_RECEIPTS.append(receipt)
-        return {"ok": True, "warning": "local_ephemeral_persistence"}
+        return {"ok": False, "code": "persistence_unavailable"}
     try:
         with _connect() as conn:
             with conn.cursor() as cur:
@@ -550,12 +558,12 @@ def _persist_comparison(
             conn.commit()
         return {"ok": True}
     except Exception:
-        return {"ok": False, "warning": "persistence_unavailable"}
+        return {"ok": False, "code": "persistence_unavailable"}
 
 
 def _load_comparison(tenant_key: str, comparison_id: str) -> dict[str, Any] | None:
     if not _database_url() or psycopg is None:
-        return _LOCAL_COMPARISONS.get((tenant_key, comparison_id))
+        return {"ok": False, "code": "persistence_unavailable"}
     try:
         with _connect() as conn:
             with conn.cursor() as cur:
@@ -577,7 +585,7 @@ def _load_comparison(tenant_key: str, comparison_id: str) -> dict[str, Any] | No
             "latest_receipt_id": row.get("latest_receipt_id"),
         }
     except Exception:
-        return None
+        return {"ok": False, "code": "persistence_unavailable"}
 
 
 def _build_receipt(
@@ -603,7 +611,7 @@ def _build_receipt(
         "schema": "quotejudge.production_receipt.v1",
         "receipt_id": receipt_id,
         "agent_name": "quote-judge-studio-v1",
-        "agent_version": "0.1.0",
+        "agent_version": AGENT_VERSION,
         "skill_name": skill_name,
         "comparison_id": comparison_id,
         "status": "ok" if result.get("ok") else "error",

frontend/package.json

@@ -1,7 +1,7 @@
 {
   "name": "quote-judge-studio-v1-frontend",
   "private": true,
-  "version": "0.1.0",
+  "version": "0.1.4",
   "type": "module",
   "scripts": {
     "dev": "vite --host 0.0.0.0 --port 5173",

requirements.txt

@@ -1,7 +1,2 @@
 # a2a-pack is auto-installed by the deploy build.
-# These starter deps power the DeepAgents tool-calling example in agent.py.
-deepagents>=0.5.0
-langchain>=0.3
-langchain-openai>=0.2
-langgraph>=0.6
 psycopg[binary]>=3.2

tests/test_full_stack_contract.py

@@ -2,12 +2,14 @@ import asyncio
 import base64
 import json
 from pathlib import Path
+from unittest.mock import Mock
 
 from a2a_pack.auth import PlatformUserAuth
 from a2a_pack.cli.local import load_local_project
 from a2a_pack.context import LocalRunContext
 
-from agent import QuoteJudgeStudioV1, _LOCAL_COMPARISONS, _LOCAL_RECEIPTS
+import agent as agent_module
+from agent import QuoteJudgeStudioV1
 
 
 ROOT = Path(__file__).resolve().parents[1]
@@ -54,11 +56,28 @@ def test_full_stack_product_contract():
     assert databases[0].migrations is not None
     assert databases[0].migrations.path == "db/migrations"
     assert card.runtime.pricing.caller_pays_llm is False
+    assert card.version == "0.1.4"
 
 
-def test_success_failure_reload_and_receipt_contract():
-    _LOCAL_COMPARISONS.clear()
-    _LOCAL_RECEIPTS.clear()
+def test_success_failure_reload_and_receipt_contract(monkeypatch):
+    comparisons = {}
+    receipts = []
+
+    def persist(tenant_key, comparison_id, result, receipt):
+        comparisons[(tenant_key, comparison_id)] = {
+            "result": json.loads(json.dumps(result)),
+            "updated_at": receipt["created_at"],
+            "latest_receipt_id": receipt["receipt_id"],
+        }
+        receipts.append(receipt)
+        return {"ok": True}
+
+    monkeypatch.setattr(agent_module, "_persist_comparison", persist)
+    monkeypatch.setattr(
+        agent_module,
+        "_load_comparison",
+        lambda tenant_key, comparison_id: comparisons.get((tenant_key, comparison_id)),
+    )
     agent = QuoteJudgeStudioV1()
     quotes = [
         {"vendor": "Acme Bearings", "quantity": 500, "unit_price": 10.75, "delivery_days": 12, "warranty_months": 12},
@@ -72,7 +91,8 @@ def test_success_failure_reload_and_receipt_contract():
     assert success["recommendation"]["vendor"] == "Beta Industrial"
     assert success["receipt"]["receipt_id"].startswith("qjr_")
     assert success["receipt"]["persisted"] is True
-    assert _LOCAL_RECEIPTS and _LOCAL_RECEIPTS[-1]["schema"] == "quotejudge.production_receipt.v1"
+    assert receipts and receipts[-1]["schema"] == "quotejudge.production_receipt.v1"
+    assert receipts[-1]["agent_version"] == "0.1.4"
     assert "tenant_key" not in json.dumps(success)
 
     reload = run(agent.local_invoke("get_comparison", auth=AUTH, comparison_id="studio-quote-v1"))
@@ -91,8 +111,8 @@ def test_success_failure_reload_and_receipt_contract():
     assert failure == {"ok": False, "code": "at_least_two_quotes_required", "comparison_id": "studio-quote-invalid"}
 
 
-def test_browser_upload_bridge_success_and_invalid_base64():
-    _LOCAL_COMPARISONS.clear()
+def test_browser_upload_bridge_success_and_invalid_base64(monkeypatch):
+    monkeypatch.setattr(agent_module, "_persist_comparison", lambda *_args: {"ok": True})
     agent = QuoteJudgeStudioV1()
     payload = base64.b64encode(json.dumps({
         "quotes": [
@@ -124,6 +144,56 @@ def test_browser_upload_bridge_success_and_invalid_base64():
     assert invalid["code"] == "invalid_base64"
 
 
+def test_persistence_fails_closed_without_managed_database(monkeypatch):
+    monkeypatch.delenv("DATABASE_URL", raising=False)
+    receipt = {"receipt_id": "qjr_test", "created_at": "2026-01-01T00:00:00+00:00"}
+
+    persisted = agent_module._persist_comparison("user:123", "quote-1", {}, receipt)
+    loaded = agent_module._load_comparison("user:123", "quote-1")
+
+    assert persisted == {"ok": False, "code": "persistence_unavailable"}
+    assert loaded == {"ok": False, "code": "persistence_unavailable"}
+
+
+def test_database_exception_returns_stable_failure_code(monkeypatch):
+    monkeypatch.setenv("DATABASE_URL", "postgresql://example.invalid/db")
+    monkeypatch.setattr(agent_module, "psycopg", Mock())
+
+    def unavailable():
+        raise TimeoutError("database connection timed out")
+
+    monkeypatch.setattr(agent_module, "_connect", unavailable)
+    persisted = agent_module._persist_comparison(
+        "user:123",
+        "quote-1",
+        {"weights": {}, "quote_count": 2, "recommendation": {"vendor": "Acme"}},
+        {
+            "receipt_id": "qjr_test",
+            "created_at": "2026-01-01T00:00:00+00:00",
+            "skill_name": "compare_quotes",
+            "status": "ok",
+            "input_hash": "input",
+            "result_hash": "result",
+        },
+    )
+
+    assert persisted == {"ok": False, "code": "persistence_unavailable"}
+
+
+def test_database_connection_has_hard_timeouts(monkeypatch):
+    fake_psycopg = Mock()
+    fake_psycopg.connect.return_value = object()
+    monkeypatch.setattr(agent_module, "psycopg", fake_psycopg)
+    monkeypatch.setenv("DATABASE_URL", "postgresql://example.invalid/db")
+
+    agent_module._connect()
+
+    _, kwargs = fake_psycopg.connect.call_args
+    assert kwargs["connect_timeout"] == 5
+    assert "statement_timeout=10000" in kwargs["options"]
+    assert "lock_timeout=5000" in kwargs["options"]
+
+
 def test_mcp_tool_schemas_are_bounded_and_typed():
     card = load_local_project(ROOT).agent_cls().card()
     skills = {skill.name: skill for skill in card.skills}