@emseepea/create-tool-server 0.0.5 → 0.0.7

package/initializer-dist/template/eval/meaning.test.mjs

@@ -1,20 +1,40 @@
-import { toolSelectionTest } from "@emseepea/testing/semantic";
+import test from "node:test";
+import {
+  assertResponseContains,
+  assertResponseMeaning,
+  assertToolCalls,
+  createConversation,
+} from "@emseepea/testing/semantic";
 
-toolSelectionTest("Pea variety details keep each concept distinct", {
-  server: new URL("../dist/server.js", import.meta.url),
-  question:
-    "For Highland Snap, give me its pea type, growth habit, days to maturity, and " +
-    "traits. Keep those concepts distinct.",
-  criticalFacts: [
-    "snap",
-    "climbing",
-    "70",
-    "edible pods",
-    "needs support"
-  ],
-  criteria:
-    "The answer says Highland Snap is a snap pea with a climbing growth habit, " +
-    "matures in 70 days, has edible pods, and needs support. It does not confuse " +
-    "pea type, growth habit, maturity time, or traits.",
-  expectedTools: ["get-pea-variety"],
+test("looks up pea varieties and compares a follow-up", async (t) => {
+  const chat = await createConversation(t, {
+    server: new URL("../dist/server.js", import.meta.url),
+  });
+
+  // The first turn's flexible prose needs the one model judge. The follow-up's
+  // exact call and returned name add context coverage without another judge.
+  const highland = await chat.send(
+    "Describe the pea type, growth habit, maturity time, and traits of Highland Snap.",
+  );
+
+  assertToolCalls(highland, [{
+    name: "get-pea-variety",
+    arguments: { name: "Highland Snap" },
+  }]);
+  assertResponseContains(highland, "Highland Snap");
+  await assertResponseMeaning(highland, {
+    expected:
+      "Highland Snap is a climbing snap pea that matures in 70 days, has edible " +
+      "pods, and needs support. Pea type, growth habit, maturity time, and traits " +
+      "remain distinct.",
+  });
+
+  const comparison = await chat.send(
+    "Compare that with Harbour Gem. Include its pea type, growth habit, and maturity time.",
+  );
+  assertToolCalls(comparison, [{
+    name: "get-pea-variety",
+    arguments: { name: "Harbour Gem" },
+  }]);
+  assertResponseContains(comparison, "Harbour Gem");
 });

package/initializer-dist/template/package.json

@@ -14,7 +14,7 @@
     "lint": "oxlint src test eval"
   },
   "devDependencies": {
-    "@emseepea/testing": "0.2.2",
+    "@emseepea/testing": "0.3.0",
     "@types/node": "24.13.3",
     "typescript": "6.0.3",
     "oxlint": "1.80.0"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@emseepea/create-tool-server",
-  "version": "0.0.5",
+  "version": "0.0.7",
   "description": "Create an Em See Pea server with one public tool.",
   "license": "MIT",
   "type": "module",
@@ -20,7 +20,7 @@
   "devDependencies": {
     "@emseepea/server": "0.3.0",
     "zod": "4.4.3",
-    "@emseepea/testing": "0.2.2",
+    "@emseepea/testing": "0.3.0",
     "@types/node": "24.13.3",
     "typescript": "6.0.3",
     "oxlint": "1.80.0"