@emseepea/create-html-ui-server 0.0.6 → 0.0.8

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

@@ -1,21 +1,38 @@
-import { toolSelectionTest } from "@emseepea/testing/semantic";
+import test from "node:test";
+import {
+  assertNoToolCalls,
+  assertResponseContains,
+  assertResponseMeaning,
+  assertToolCalls,
+  createConversation,
+} from "@emseepea/testing/semantic";
 
-toolSelectionTest("Native UI preview is not mistaken for a completed effect", {
-  server: new URL("../dist/server.js", import.meta.url),
-  question:
-    "Summarize the snap-pea planting-plan preview. Was a report sent or stored, and did this " +
-    "operation change anything?",
-  criticalFacts: [
-    "Highland Snap",
-    "Meadow Sweet",
-    "preview-only",
-    "false",
-    /\b(?:no|not|nothing|neither|never|wasn't|weren't|didn't)\b[^.!?\n]{0,80}\b(?:sent|delivered)\b/i,
-    /\b(?:no|not|nothing|neither|never|wasn't|weren't|didn't)\b[^.!?\n]{0,80}\b(?:stored|saved|persisted)\b/i
-  ],
-  criteria:
-    "The answer identifies Highland Snap and Meadow Sweet as the two snap-pea " +
-    "matches. It says the result is preview-only, effectPerformed is false, no report " +
-    "was sent or stored, and no external action or data change occurred.",
-  expectedTools: ["preview-planting-plan"],
+test("does not mistake a native UI preview for a completed effect", async (t) => {
+  const chat = await createConversation(t, {
+    server: new URL("../dist/server.js", import.meta.url),
+  });
+
+  // The judged turn covers the UI tool's main semantic risk: preview is not an
+  // effect. A literal follow-up adds conversation coverage without another judge.
+  const response = await chat.send(
+    "Preview a plan titled Snap pea plan for snap peas, including growing tips. " +
+    "Summarize the matches and say whether anything was sent, stored, or changed.",
+  );
+
+  assertToolCalls(response, [{
+    name: "preview-planting-plan",
+    arguments: { title: "Snap pea plan", peaType: "snap", includeTips: true },
+  }]);
+  assertResponseContains(response, "Highland Snap");
+  await assertResponseMeaning(response, {
+    expected:
+      "The preview contains Highland Snap and Meadow Sweet. It is preview-only, " +
+      "performed no effect, sent and stored no report, and changed no external data.",
+  });
+
+  const followUp = await chat.send(
+    "How many snap varieties were in that preview? Reply with the numeral only.",
+  );
+  assertNoToolCalls(followUp);
+  assertResponseContains(followUp, "2");
 });

package/initializer-dist/template/package.json

@@ -16,7 +16,7 @@
   "devDependencies": {
     "playwright": "1.62.1",
     "axe-core": "4.13.0",
-    "@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-html-ui-server",
-  "version": "0.0.6",
+  "version": "0.0.8",
   "description": "Create an Em See Pea server with an accessible HTML form.",
   "license": "MIT",
   "type": "module",
@@ -23,7 +23,7 @@
     "@emseepea/example-ui-shared": "0.0.0",
     "playwright": "1.62.1",
     "axe-core": "4.13.0",
-    "@emseepea/testing": "0.2.2",
+    "@emseepea/testing": "0.3.0",
     "@types/node": "24.13.3",
     "typescript": "6.0.3",
     "oxlint": "1.80.0"