test(stories): add sample conversations for the chat stories

Add deterministic samples built only from the conversation contract:
a short chat, a 300-message history, a tool storm with a batch, a
failure and every approval outcome, generated images, a thread and a
failed send. playTurn reveals a message's parts over time for streaming
stories. Images are inline SVG, timestamps are fixed, and nothing reads
the clock or Math.random.

Author: pratikbuilds

stories/chat/samples/conversations.test.ts

@@ -0,0 +1,159 @@
+import { describe, expect, test } from "bun:test";
+
+import { isAwaitingApproval } from "../../../src/lib/conversation-activity.js";
+import type {
+  ConversationMessage,
+  ConversationPart,
+  PartTool,
+} from "../../../src/lib/conversation.js";
+import {
+  SAMPLE_NOW,
+  failedSend,
+  generatedImages,
+  longChat,
+  samples,
+  toolStorm,
+  withThread,
+} from "./conversations.js";
+
+function everyMessage(sample: (typeof samples)[number]) {
+  return [...sample.messages, ...(sample.thread?.replies ?? [])];
+}
+
+function toolsOf(parts: readonly ConversationPart[]): readonly PartTool[] {
+  return parts.flatMap((part) => {
+    if (part.kind === "tool-trace") return [part];
+    if (part.kind === "tool-batch") return part.tools;
+    return [];
+  });
+}
+
+function allParts(messages: readonly ConversationMessage[]) {
+  return messages.flatMap((message) => message.parts);
+}
+
+describe.each(samples.map((sample) => [sample.title, sample] as const))(
+  "%s",
+  (_, sample) => {
+    test("has unique message ids, thread included", () => {
+      const ids = everyMessage(sample).map((message) => message.id);
+      expect(new Set(ids).size).toBe(ids.length);
+    });
+
+    test("never goes back in time and stays before SAMPLE_NOW", () => {
+      for (const list of [sample.messages, sample.thread?.replies ?? []]) {
+        const times = list.map((message) => Date.parse(message.at));
+        expect(times).toEqual([...times].sort((a, b) => a - b));
+        for (const time of times) {
+          expect(time).toBeLessThan(Date.parse(SAMPLE_NOW));
+        }
+      }
+    });
+
+    test("uses generic tool names and example.com sources", () => {
+      const parts = allParts(everyMessage(sample));
+      for (const tool of toolsOf(parts)) {
+        expect(tool.name).toMatch(
+          /^(runs|files|web|calendar|notes|mail|records|image)__[a-z_]+$/,
+        );
+      }
+      for (const part of parts) {
+        if (part.kind !== "sources") continue;
+        for (const source of part.sources) {
+          expect(new URL(source.url).hostname).toBe("example.com");
+        }
+      }
+    });
+
+    test("gives every image alt text and keeps it off the network", () => {
+      for (const part of allParts(sample.messages)) {
+        if (part.kind !== "image") continue;
+        for (const image of part.images) {
+          expect(image.alt).toBeTruthy();
+          if (image.url !== undefined) {
+            expect(image.url.startsWith("data:")).toBe(true);
+          }
+        }
+      }
+    });
+  },
+);
+
+test("samples read no randomness and no clock", async () => {
+  const source = await Bun.file(
+    new URL("./conversations.ts", import.meta.url),
+  ).text();
+  expect(source).not.toMatch(/Math\.random|Date\.now|new Date\(\)/);
+});
+
+test("longChat holds a few hundred messages across more than one day", () => {
+  expect(longChat.messages.length).toBeGreaterThanOrEqual(200);
+  const days = new Set(
+    longChat.messages.map((message) => message.at.slice(0, 10)),
+  );
+  expect(days.size).toBeGreaterThan(1);
+});
+
+test("toolStorm covers a batch, a failure and every approval outcome", () => {
+  const parts = allParts(toolStorm.messages);
+  expect(parts.some((part) => part.kind === "tool-batch")).toBe(true);
+  const tools = toolsOf(parts);
+  expect(tools.length).toBeGreaterThan(40);
+  expect(tools.some((tool) => tool.status === "error")).toBe(true);
+  expect(new Set(tools.map((tool) => tool.approval?.status))).toEqual(
+    new Set([
+      undefined,
+      "pending",
+      "approved",
+      "rejected",
+      "timeout",
+      "expired",
+    ]),
+  );
+  expect(
+    new Set(tools.map((tool) => tool.approval?.scope).filter(Boolean)),
+  ).toEqual(new Set(["once", "always"]));
+  expect(tools.filter(isAwaitingApproval)).toHaveLength(1);
+});
+
+test("toolStorm's calls that never ran carry no output", () => {
+  for (const tool of toolsOf(allParts(toolStorm.messages))) {
+    if (tool.status === "output-denied") {
+      expect(tool.output).toBeUndefined();
+    }
+  }
+});
+
+test("generatedImages covers both storage forms and every image state", () => {
+  const images = allParts(generatedImages.messages).flatMap((part) =>
+    part.kind === "image" ? [part] : [],
+  );
+  expect(new Set(images.map((part) => part.state))).toEqual(
+    new Set(["done", "generating", "error"]),
+  );
+  const stored = images.flatMap((part) => part.images);
+  expect(stored.some((image) => image.base64 !== undefined)).toBe(true);
+  expect(stored.some((image) => image.url !== undefined)).toBe(true);
+  const generating = images.find((part) => part.state === "generating");
+  expect(generating?.expected?.count).toBeGreaterThan(0);
+});
+
+test("withThread's reply summary matches its thread", () => {
+  const thread = withThread.thread;
+  if (thread === undefined) throw new Error("withThread has no thread");
+  const parent = withThread.messages.find(
+    (message) => message.id === thread.parentId,
+  );
+  expect(parent?.replies).toEqual({
+    count: thread.replies.length,
+    lastAt: thread.replies.at(-1)?.at ?? "",
+  });
+});
+
+test("failedSend has a failed send with a reason and one still sending", () => {
+  const failed = failedSend.messages.find(
+    (message) => message.status === "failed",
+  );
+  expect(failed?.error).toBeTruthy();
+  expect(failedSend.messages.some((m) => m.status === "sending")).toBe(true);
+});