@sleep2agi/commhub-server 0.9.0-preview.4 → 0.9.0-preview.40

package/src/delivered-without-push.test.ts

@@ -0,0 +1,140 @@
+// f015d9d6 症状①：tasks.status 被置为 'delivered'，而 SSE 推送根本没发出。
+//
+// send_task 无条件以字面量 'delivered' 插入 tasks 行，推送却另外被
+// `if (target.state === "online")` 挡着（tools.ts:1432）。同一段代码里给
+// network observer 的事件把同一件事标成 "queued" —— 也就是说代码**知道**
+// 它没投出去，却仍然把 delivered 写进了 tasks.status。
+//
+// 而 "online" 的判据是 resolveDeliveryTarget 里的
+//   stale = now - (last_seen_at || updated_at) > 5 分钟
+// 且 last_seen_at 只在 report_status 时刷新（tools.ts:674/702）。
+// 全网「停发心跳/状态类消息」的省额度规则之下，一个安静的 MCP 会话必然变 stale。
+import { afterAll, beforeEach, describe, expect, test } from "bun:test";
+import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
+import { db } from "./db.js";
+import { registerTools } from "./tools.js";
+import { __resetSSEClientsForTest, createSSEStream } from "./push.js";
+
+const NET = "net_delivered_push_f015";
+const USER = "user_delivered_push_f015";
+const SENDER = "sender-f015b";
+const TARGET = "quiet-mcp-f015b";
+
+type ToolHandler = (args: any) => Promise<{ content: Array<{ type: "text"; text: string }> }>;
+
+function cleanup() {
+  __resetSSEClientsForTest();
+  for (const t of ["tasks", "inbox", "task_events", "sessions", "api_tokens", "nodes", "rename_txn"]) {
+    try { db.run(`DELETE FROM ${t} WHERE network_id = ?1`, [NET]); } catch {}
+  }
+  try { db.run("DELETE FROM network_members WHERE network_id = ?1", [NET]); } catch {}
+  try { db.run("DELETE FROM networks WHERE network_id = ?1", [NET]); } catch {}
+  try { db.run("DELETE FROM users WHERE user_id = ?1", [USER]); } catch {}
+}
+
+/** minutesAgo 控制目标会话的 last_seen_at —— 这是本文件唯一要动的变量。 */
+function seed(minutesAgo: number) {
+  db.run("INSERT INTO users (user_id, username, password_hash, role, created_at) VALUES (?1, ?1, 'x', 'admin', datetime('now'))", [USER]);
+  db.run("INSERT INTO networks (network_id, network_name, owner_id, created_at) VALUES (?1, ?1, ?2, datetime('now'))", [NET, USER]);
+  db.run("INSERT INTO network_members (user_id, network_id, role, joined_at) VALUES (?1, ?2, 'owner', datetime('now'))", [USER, NET]);
+  for (const alias of [SENDER, TARGET]) {
+    db.run(
+      "INSERT INTO api_tokens (token_id, token_hash, user_id, network_id, name, scope) VALUES (?1, ?2, ?3, ?4, ?5, 'network')",
+      [`token-${alias}`, `hash-${alias}`, USER, NET, `node:${alias}`],
+    );
+  }
+  db.run(
+    "INSERT INTO sessions (resume_id, alias, status, node_id, network_id, updated_at, last_seen_at) VALUES (?1, ?1, 'idle', NULL, ?2, datetime('now'), datetime('now', ?3))",
+    [SENDER, NET, "-1 minutes"],
+  );
+  // 🔴 目标：MCP 挂载的 claude-code 会话形状（node_id NULL），
+  //    status 是 'idle'（不是 offline），只是安静了 minutesAgo 分钟。
+  db.run(
+    "INSERT INTO sessions (resume_id, alias, status, node_id, network_id, updated_at, last_seen_at) VALUES (?1, ?2, 'idle', NULL, ?3, datetime('now', ?4), datetime('now', ?4))",
+    [`cc-n_${TARGET}`, TARGET, NET, `-${minutesAgo} minutes`],
+  );
+}
+
+function toolsFor(alias: string): Record<string, ToolHandler> {
+  const server = new McpServer({ name: "delivered-push-f015", version: "0" }) as any;
+  const tools: Record<string, ToolHandler> = {};
+  const original = server.tool.bind(server);
+  server.tool = (name: string, description: string, schema: any, handler: ToolHandler) => {
+    tools[name] = handler; return original(name, description, schema, handler);
+  };
+  registerTools(server, undefined, NET, USER, alias, true, `token-${alias}`);
+  return tools;
+}
+
+/** 从订阅者那一侧真正读出字节 —— 地面真相，不是连接计数。 */
+async function readUntil(reader: ReadableStreamDefaultReader<Uint8Array>, needle: string, ms = 1500): Promise<string | null> {
+  const dec = new TextDecoder();
+  let buf = "";
+  const deadline = Date.now() + ms;
+  while (Date.now() < deadline) {
+    const chunk = await Promise.race([
+      reader.read(),
+      new Promise<{ done: true; value: undefined }>((r) => setTimeout(() => r({ done: true, value: undefined }), 200)),
+    ]);
+    if (chunk.value) buf += dec.decode(chunk.value, { stream: true });
+    if (buf.includes(needle)) return buf;
+  }
+  return null;
+}
+
+let sendSeq = 0;
+/** minutesAgo = 目标安静了多久；withSubscriber = 是否真的有人连着。 */
+async function send(minutesAgo: number, withSubscriber = true) {
+  cleanup(); seed(minutesAgo);
+  let reader: ReadableStreamDefaultReader<Uint8Array> | null = null;
+  if (withSubscriber) {
+    const res = createSSEStream(TARGET, NET);
+    reader = res.body!.getReader();
+  }
+  const r = await toolsFor(SENDER).send_task({ alias: TARGET, task: `ping-${++sendSeq}`, priority: "normal" });
+  const out = JSON.parse(r.content[0].text);
+  const row = db.get<{ status: string }>("SELECT status FROM tasks WHERE to_name = ?1 AND network_id = ?2", TARGET, NET);
+  const received = reader ? await readUntil(reader, "new_task") : null;
+  return { out, taskStatus: row?.status ?? null, received };
+}
+
+describe("f015d9d6 症状① — delivered 置位与实际推送不同步", () => {
+  beforeEach(() => cleanup());
+  afterAll(() => cleanup());
+
+  test("对照：目标 1 分钟前还活跃 ⇒ 订阅者收到 new_task", async () => {
+    const { taskStatus, received } = await send(1);
+    expect(taskStatus).toBe("delivered");
+    expect(received).toContain("new_task");
+  });
+
+  test("🔴 核心：目标安静 10 分钟但订阅者连着 ⇒ 仍然收得到（修复前这里收到 null）", async () => {
+    const { taskStatus, received } = await send(10);
+    expect(received).toContain("new_task");
+    expect(taskStatus).toBe("delivered");   // 现在这个 delivered 是真话了
+  });
+
+  test("🔴 可达性判据必须来自订阅者注册表，不是 last_seen_at 的时间戳猜测", async () => {
+    // 安静 10 分钟 = 时间戳判据会说 offline；而实际上有人在听。
+    const quiet = await send(10);
+    expect(quiet.received).toContain("new_task");
+    // 真没人听的时候，pushEvent 本来就是 no-op，不需要额外的闸来保护。
+    const nobody = await send(10, false);
+    expect(nobody.received).toBeNull();
+  });
+
+  test("真的没有订阅者时，tasks 行仍写 delivered —— 本 PR 刻意不改，记录在案", async () => {
+    const { taskStatus, received } = await send(10, false);
+    expect(received).toBeNull();
+    // 这一条钉的是**当前行为**，不是我认为对的行为：没人收到，行仍自称 delivered。
+    expect(taskStatus).toBe("delivered");
+  });
+
+  test("活性判据的输入 last_seen_at 只由 report_status 刷新 —— 发送方不刷新它", async () => {
+    cleanup(); seed(4);
+    const before = db.get<{ last_seen_at: string }>("SELECT last_seen_at FROM sessions WHERE alias = ?1 AND network_id = ?2", TARGET, NET);
+    await toolsFor(SENDER).send_task({ alias: TARGET, task: `ping-${++sendSeq}`, priority: "normal" });
+    const after = db.get<{ last_seen_at: string }>("SELECT last_seen_at FROM sessions WHERE alias = ?1 AND network_id = ?2", TARGET, NET);
+    expect(after?.last_seen_at).toBe(before!.last_seen_at);
+  });
+});

package/src/desktop-message.test.ts

@@ -0,0 +1,286 @@
+// Desktop user push contract.
+//
+// Pins the new user-keyed SSE path and send_desktop_message MCP tool:
+// - user streams are keyed by networkId:userId, never alias
+// - ntok callers cannot cross network by supplying a foreign to_user_id/network_id
+// - target ambiguity/miss is fail-closed before audit/push
+
+import { afterAll, afterEach, beforeEach, describe, expect, test } from "bun:test";
+import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
+import { db } from "./db.js";
+import { registerTools } from "./tools.js";
+import {
+  __resetSSEClientsForTest,
+  createUserEventStream,
+  pushUserEvent,
+} from "./push.js";
+
+const NET_A = "net_desktop_a";
+const NET_B = "net_desktop_b";
+const NODE_USER = "u_desktop_node_owner";
+const SENDER_USER = "u_desktop_sender";
+const TARGET_A = "u_desktop_target_a";
+const TARGET_B = "u_desktop_target_b";
+const TARGET_BOTH = "u_desktop_target_both";
+const ALL_USERS = [NODE_USER, SENDER_USER, TARGET_A, TARGET_B, TARGET_BOTH];
+
+type ToolHandler = (args: any) => Promise<{ content: Array<{ type: "text"; text: string }> }>;
+
+function cleanup() {
+  try { db.run("DELETE FROM audit_log WHERE action = 'send_desktop_message' OR network_id IN (?1, ?2)", [NET_A, NET_B]); } catch {}
+  try { db.run("DELETE FROM network_members WHERE network_id IN (?1, ?2)", [NET_A, NET_B]); } catch {}
+  try { db.run("DELETE FROM networks WHERE network_id IN (?1, ?2)", [NET_A, NET_B]); } catch {}
+  for (const u of ALL_USERS) {
+    try { db.run("DELETE FROM users WHERE user_id = ?1", [u]); } catch {}
+  }
+}
+
+function seed() {
+  for (const u of ALL_USERS) {
+    db.run(
+      `INSERT INTO users (user_id, username, password_hash, role, created_at)
+       VALUES (?1, ?2, 'x', 'user', datetime('now'))`,
+      [u, `${u}_name`],
+    );
+  }
+  db.run(`INSERT INTO networks (network_id, network_name, owner_id, created_at) VALUES (?1, ?1, ?2, datetime('now'))`, [NET_A, SENDER_USER]);
+  db.run(`INSERT INTO networks (network_id, network_name, owner_id, created_at) VALUES (?1, ?1, ?2, datetime('now'))`, [NET_B, TARGET_B]);
+  const member = (userId: string, networkId: string, role: string) =>
+    db.run(`INSERT INTO network_members (user_id, network_id, role, joined_at) VALUES (?1, ?2, ?3, datetime('now'))`, [userId, networkId, role]);
+  member(NODE_USER, NET_A, "member");
+  member(SENDER_USER, NET_A, "owner");
+  member(TARGET_A, NET_A, "member");
+  member(TARGET_B, NET_B, "member");
+  member(TARGET_BOTH, NET_A, "member");
+  member(TARGET_BOTH, NET_B, "member");
+}
+
+beforeEach(() => { cleanup(); seed(); });
+afterEach(() => { __resetSSEClientsForTest(); });
+afterAll(cleanup);
+
+function buildHandlers(opts: { netId?: string | null; userId?: string | null; alias?: string | null; isNetwork?: boolean }) {
+  const server = new McpServer({ name: "test", version: "0" }) as any;
+  const tools: Record<string, ToolHandler> = {};
+  const origTool = server.tool.bind(server);
+  server.tool = (name: string, _desc: string, schema: any, handler: ToolHandler) => {
+    tools[name] = handler;
+    return origTool(name, _desc, schema, handler);
+  };
+  registerTools(
+    server,
+    undefined,
+    opts.netId ?? null,
+    opts.userId ?? null,
+    opts.alias ?? opts.userId ?? null,
+    opts.isNetwork ?? false,
+    "tok_desktop_test",
+  );
+  return tools;
+}
+
+async function call(handler: ToolHandler, args: any): Promise<Record<string, any>> {
+  const r = await handler(args);
+  return JSON.parse(r.content[0].text);
+}
+
+async function readFrame(
+  reader: ReadableStreamDefaultReader<Uint8Array>,
+  timeoutMs = 2_000,
+): Promise<Record<string, any>> {
+  const decoder = new TextDecoder();
+  let buf = "";
+  const deadline = Date.now() + timeoutMs;
+  while (Date.now() < deadline) {
+    const result = await Promise.race([
+      reader.read(),
+      new Promise<never>((_, rej) =>
+        setTimeout(() => rej(new Error(`no SSE frame within ${timeoutMs}ms`)), deadline - Date.now()),
+      ),
+    ]);
+    if (result.done) throw new Error("SSE stream ended unexpectedly");
+    buf += decoder.decode(result.value, { stream: true });
+    const sep = buf.indexOf("\n\n");
+    if (sep === -1) continue;
+    const rawFrame = buf.slice(0, sep);
+    buf = buf.slice(sep + 2);
+    const dataLine = rawFrame.split("\n").find((l) => l.startsWith("data: "));
+    if (!dataLine) continue;
+    return JSON.parse(dataLine.slice(6));
+  }
+  throw new Error(`no SSE frame within ${timeoutMs}ms`);
+}
+
+function auditCount() {
+  return db.get<{ n: number }>("SELECT COUNT(*) AS n FROM audit_log WHERE action = 'send_desktop_message'")?.n ?? 0;
+}
+
+describe("send_desktop_message", () => {
+  test("ntok caller can push to a user in the same bound network", async () => {
+    const stream = createUserEventStream(NET_A, TARGET_A);
+    const reader = stream.body!.getReader();
+    await readFrame(reader);
+
+    const tools = buildHandlers({ netId: NET_A, userId: NODE_USER, alias: "node-a", isNetwork: true });
+    const result = await call(tools.send_desktop_message, {
+      to_user_id: TARGET_A,
+      message: "desktop hello",
+      severity: "success",
+    });
+
+    expect(result.ok).toBe(true);
+    expect(result.message_id).toStartWith("dm_");
+    expect(auditCount()).toBe(1);
+    const evt = await readFrame(reader);
+    expect(evt.type).toBe("desktop_message");
+    expect(evt.message_id).toBe(result.message_id);
+    expect(evt.message).toBe("desktop hello");
+    expect(evt.from).toBe("node-a");
+    expect(evt.network_id).toBe(NET_A);
+    expect(evt.user_id).toBe(TARGET_A);
+    expect(evt.scope).toBe("user");
+    await reader.cancel();
+  });
+
+  test("utok caller can push with explicit network_id to a member user", async () => {
+    const stream = createUserEventStream(NET_A, TARGET_A);
+    const reader = stream.body!.getReader();
+    await readFrame(reader);
+
+    const tools = buildHandlers({ userId: SENDER_USER, alias: "sender-user" });
+    const result = await call(tools.send_desktop_message, {
+      to_username: `${TARGET_A}_name`,
+      network_id: NET_A,
+      title: "Heads up",
+      message: "from user token",
+    });
+
+    expect(result.ok).toBe(true);
+    const evt = await readFrame(reader);
+    expect(evt.title).toBe("Heads up");
+    expect(evt.message).toBe("from user token");
+    expect(evt.network_id).toBe(NET_A);
+    await reader.cancel();
+  });
+
+  test("ntok caller cannot cross network with foreign network_id and to_user_id; no audit or push happens first", async () => {
+    const stream = createUserEventStream(NET_B, TARGET_B);
+    const reader = stream.body!.getReader();
+    await readFrame(reader);
+
+    const tools = buildHandlers({ netId: NET_A, userId: NODE_USER, alias: "node-a", isNetwork: true });
+    const result = await call(tools.send_desktop_message, {
+      to_user_id: TARGET_B,
+      network_id: NET_B,
+      message: "must not cross",
+    });
+
+    expect(result).toEqual({ ok: false, error: "desktop_target_not_in_network", network_id: NET_A });
+    expect(auditCount()).toBe(0);
+
+    pushUserEvent(NET_B, TARGET_B, { type: "desktop_message", message_id: "marker_after_denied" });
+    expect((await readFrame(reader)).message_id).toBe("marker_after_denied");
+    await reader.cancel();
+  });
+
+  test("missing target is fail-closed before audit", async () => {
+    const tools = buildHandlers({ netId: NET_A, userId: NODE_USER, alias: "node-a", isNetwork: true });
+    const result = await call(tools.send_desktop_message, { message: "nobody" });
+    expect(result).toEqual({
+      ok: false,
+      error: "desktop_target_required",
+      message: "to_user_id or to_username is required",
+    });
+    expect(auditCount()).toBe(0);
+  });
+
+  test("conflicting to_user_id and to_username is fail-closed before audit", async () => {
+    const tools = buildHandlers({ netId: NET_A, userId: NODE_USER, alias: "node-a", isNetwork: true });
+    const result = await call(tools.send_desktop_message, {
+      to_user_id: TARGET_A,
+      to_username: `${TARGET_BOTH}_name`,
+      message: "conflict",
+    });
+    expect(result).toEqual({
+      ok: false,
+      error: "desktop_target_mismatch",
+      to_user_id: TARGET_A,
+      to_username: `${TARGET_BOTH}_name`,
+    });
+    expect(auditCount()).toBe(0);
+  });
+
+  test("unknown target is fail-closed before audit", async () => {
+    const tools = buildHandlers({ netId: NET_A, userId: NODE_USER, alias: "node-a", isNetwork: true });
+    const result = await call(tools.send_desktop_message, {
+      to_username: "not_a_real_user",
+      message: "lost",
+    });
+    expect(result).toEqual({ ok: false, error: "desktop_target_not_found", field: "to_username" });
+    expect(auditCount()).toBe(0);
+  });
+});
+
+// #1459 —— 返回值必须反映真实投递态。
+//
+// send_desktop_message 是 fire-and-forget：只 INSERT audit_log（不写 inbox），
+// 再 pushUserEvent 扇出；而 pushUserEvent 在没有订阅者时静默 return。
+// 用户 dashboard 离线/重连时消息永久丢（无 DB 行 ⇒ 无从 reconcile，
+// /api/messages 只 SELECT ... FROM inbox），而调用方拿到的是 ok:true。
+//
+// 本 PR 只做「让丢失可见」这一半：不碰 schema，返回值说实话。
+// 持久化 + 重连补投（让丢失不发生）是另一条，方向已定为按 user_id 寻址的新表。
+describe("#1459 send_desktop_message 返回值反映真实投递态", () => {
+  test("🔴 无活跃订阅者 ⇒ delivered:false + 可判读的 reason，而不是笼统 ok:true", async () => {
+    // 故意不建 SSE 流：这正是用户 dashboard 关着的情形
+    const tools = buildHandlers({ netId: NET_A, userId: NODE_USER, alias: "node-a", isNetwork: true });
+    const before = auditCount();
+    const result = await call(tools.send_desktop_message, {
+      to_user_id: TARGET_A,
+      message: "nobody is listening",
+      severity: "info",
+    });
+
+    // ok 仍为 true：调用被接受、审计已落库 —— 这一格没变
+    expect(result.ok).toBe(true);
+    expect(result.message_id).toStartWith("dm_");
+    expect(auditCount()).toBe(before + 1);
+
+    // 新增的那一格：投递态
+    expect(result.delivered).toBe(false);
+    expect(result.reason).toBe("no_live_subscriber");
+  });
+
+  test("有活跃订阅者 ⇒ delivered:true 且不带 reason", async () => {
+    const stream = createUserEventStream(NET_A, TARGET_A);
+    const reader = stream.body!.getReader();
+    await readFrame(reader);
+
+    const tools = buildHandlers({ netId: NET_A, userId: NODE_USER, alias: "node-a", isNetwork: true });
+    const result = await call(tools.send_desktop_message, {
+      to_user_id: TARGET_A,
+      message: "someone is listening",
+      severity: "info",
+    });
+
+    expect(result.delivered).toBe(true);
+    expect(result.reason).toBeUndefined();
+    // 前提断言：这一条真的送到了订阅者手上，否则 delivered:true 就是空话
+    const evt = await readFrame(reader);
+    expect(evt.message).toBe("someone is listening");
+    await reader.cancel();
+  });
+
+  test("🔴 两种情形必须给出不同读数 —— 防止判据退化成恒真", async () => {
+    const tools = buildHandlers({ netId: NET_A, userId: NODE_USER, alias: "node-a", isNetwork: true });
+    const offline = await call(tools.send_desktop_message, { to_user_id: TARGET_A, message: "off", severity: "info" });
+
+    const stream = createUserEventStream(NET_A, TARGET_A);
+    const reader = stream.body!.getReader();
+    await readFrame(reader);
+    const online = await call(tools.send_desktop_message, { to_user_id: TARGET_A, message: "on", severity: "info" });
+    await reader.cancel();
+
+    expect(offline.delivered).not.toBe(online.delivered);
+  });
+});

package/src/ext-token-shared.test.ts

@@ -0,0 +1,60 @@
+import { describe, expect, test } from "bun:test";
+import {
+  buildStoragePath,
+  pathForExistingBlob,
+  validateIndexEntry,
+} from "./uploads.js";
+
+const fileId = "0123456789abcdef0123456789abcdef";
+const dateBucket = "2026-08-09";
+const uploadsRoot = "/srv/uploads";
+
+const malicious = [
+  "../x",
+  "/abs/x",
+  ".png/../../y",
+  ".p ng",
+  ".p\0ng",
+  ".\\..\\x",
+  "." + "a".repeat(17),
+  ".tar.gz",
+];
+
+const legal = ["", ".png", ".PNG", ".mp4", "." + "a".repeat(16)];
+
+describe("#527 shared extension-token invariant", () => {
+  test("new-upload path rejects every malformed token and accepts the shared legal set", () => {
+    for (const ext of malicious) {
+      expect(() => buildStoragePath(fileId, ext, { uploadsRoot })).toThrow(/ext .* is invalid/);
+    }
+    for (const ext of legal) {
+      expect(buildStoragePath(fileId, ext, { uploadsRoot }).ext).toBe(ext);
+    }
+  });
+
+  test("existing-blob path rejects every malformed token and accepts the shared legal set", () => {
+    for (const ext of malicious) {
+      expect(() => pathForExistingBlob(dateBucket, fileId, ext, { uploadsRoot })).toThrow(/ext .* is invalid/);
+    }
+    for (const ext of legal) {
+      expect(pathForExistingBlob(dateBucket, fileId, ext, { uploadsRoot }).ext).toBe(ext);
+    }
+  });
+
+  test("stored-index gate rejects every malformed token and accepts the shared legal set", () => {
+    for (const ext of malicious) {
+      expect(validateIndexEntry({ file_id: fileId, date_bucket: dateBucket, ext, size: 0 })).toBe(false);
+    }
+    for (const ext of legal) {
+      expect(validateIndexEntry({ file_id: fileId, date_bucket: dateBucket, ext, size: 0 })).toBe(true);
+    }
+  });
+
+  test("all three boundaries fail closed on non-string runtime values", () => {
+    for (const ext of [undefined, null, 7, {}, []]) {
+      expect(() => buildStoragePath(fileId, ext as never, { uploadsRoot })).toThrow(/ext .* is invalid/);
+      expect(() => pathForExistingBlob(dateBucket, fileId, ext as never, { uploadsRoot })).toThrow(/ext .* is invalid/);
+      expect(validateIndexEntry({ file_id: fileId, date_bucket: dateBucket, ext, size: 0 })).toBe(false);
+    }
+  });
+});

package/src/ext-validation-regression.test.ts

@@ -0,0 +1,131 @@
+// Task #25 — regression pins for ext validation in validateIndexEntry.
+//
+// 🔴 SCOPE: this suite exercises the stored-index boundary through
+// `validateIndexEntry`. Issue #527 removed the three independent copies
+// of the stored extension-token grammar: buildStoragePath,
+// pathForExistingBlob and validateIndexEntry now use one shared private
+// validator. `ext-token-shared.test.ts` separately pins all three call
+// sites. sanitizeExt remains intentionally independent because it parses
+// a client filename rather than validating an already-extracted token.
+//
+// Background (通信龙 d01bb8ce): the stored extension-token grammar
+// `/^\.[A-Za-z0-9]{1,16}$/` correctly rejects 11 known malicious ext
+// values (path-escape, traversal, absolute paths, doubles, NUL bytes,
+// backslash forms, over-length, whitespace). Verified via real-run on
+// origin/main by lead — 11/11 malicious values → false.
+//
+// But NO test locked that behavior. Anyone who widens the regex in the
+// future (accidentally or on purpose) would not hit a red test — the
+// live protection would degrade silently. This suite pins current
+// behavior in two directions:
+//
+//   1. SAFETY  (欠修 direction) — 11 malicious values → false, one row
+//      per value with the exact literal. If regex is widened to accept
+//      any of them, that row turns red.
+//   2. NON-REGRESSION (过修 direction) — a set of common legitimate
+//      extensions → true. If regex is tightened to reject any of them
+//      (e.g. someone locks down to a whitelist), that row turns red.
+//
+// Witnessed-red for the SAFETY rows: temporarily widen the regex to
+// `/./` (or comment it out) in validateIndexEntry, run this suite,
+// confirm all 11 malicious rows turn red, then revert. The mutation
+// itself is not committed; the assertion IS the mutation-red proof
+// for the reader. Recording the mutation output in the followup PR
+// body (per Constraint 3 discipline established in #503).
+//
+// Pure unit test — no server setup, no DB, no filesystem. The regex
+// lives in a pure function, so the guard is testable at that level.
+
+import { describe, expect, test } from "bun:test";
+import { validateIndexEntry } from "./uploads.js";
+
+// A structurally-valid base entry — file_id / date_bucket / size all
+// pass their own checks — so `ext` is the only field a test row varies.
+// This isolates the ext regex as the deciding factor.
+const validBase = {
+  file_id: "0123456789abcdef0123456789abcdef",
+  date_bucket: "2026-07-30",
+  size: 0,
+};
+
+describe("Task #25 (#503 followup) — ext validation regression pins", () => {
+  describe("SAFETY: the 11 malicious ext values from lead d01bb8ce all return false", () => {
+    // Each row uses the exact literal from lead's verification list so
+    // a widening regex catches on the same value the manual audit did.
+    const malicious: Array<{ label: string; ext: string }> = [
+      { label: "path-escape prefix",          ext: "../x" },
+      { label: "deep traversal to /etc",       ext: "../../etc/passwd" },
+      { label: "absolute path",               ext: "/abs/x" },
+      { label: "double-dot alone",            ext: ".." },
+      { label: "single-dot alone",            ext: "." },
+      { label: "chained after legit prefix",  ext: ".png/../../y" },
+      { label: "embedded whitespace",         ext: ".p ng" },
+      { label: "over-length (>16 chars)",     ext: "." + "a".repeat(17) },
+      { label: "embedded NUL byte",           ext: ".p\x00ng" },
+      { label: "backslash traversal",         ext: ".\\..\\x" },
+      { label: "double-backslash prefix",     ext: "..\\x" },
+    ];
+    for (const { label, ext } of malicious) {
+      test(`rejects: ${label} — ext=${JSON.stringify(ext)}`, () => {
+        expect(validateIndexEntry({ ...validBase, ext })).toBe(false);
+      });
+    }
+
+    test("precondition: 11 malicious cases enumerated (defends against silent test-shrink)", () => {
+      // If someone deletes rows from the malicious list, this fails
+      // instead of the deletion silently narrowing the guard. Number
+      // matches lead's verification list — bump only when lead adds a
+      // case and confirms it via real-run.
+      expect(malicious.length).toBe(11);
+    });
+  });
+
+  describe("NON-REGRESSION: common legitimate extensions all return true (防过修)", () => {
+    // These are the extensions the live upload path actually produces
+    // via sanitizeExt / mime-typed multipart uploads. If someone
+    // tightens the regex to a stricter whitelist, any of these turning
+    // red flags a break in a normal user path.
+    const legal = [".png", ".pdf", ".pptx", ".jpg", ".jpeg", ".txt", ".mp4", ".zip", ".json", ".md"];
+    for (const ext of legal) {
+      test(`accepts: ext=${JSON.stringify(ext)}`, () => {
+        expect(validateIndexEntry({ ...validBase, ext })).toBe(true);
+      });
+    }
+
+    test("empty ext is accepted (file with no extension is a legitimate upload)", () => {
+      expect(validateIndexEntry({ ...validBase, ext: "" })).toBe(true);
+    });
+
+    test("precondition: at least 10 legitimate ext values pinned", () => {
+      // Same discipline as the malicious side — reject silent shrink.
+      expect(legal.length).toBeGreaterThanOrEqual(10);
+    });
+  });
+
+  describe("regex shape assumptions (documented, not enforced by the function)", () => {
+    // These rows document facts the current regex enforces beyond the
+    // 11 malicious literals. They are close to the boundary and would
+    // be the first things to move if someone reworked the regex.
+    test("upper-case extensions accepted (regex is case-insensitive by [A-Za-z])", () => {
+      expect(validateIndexEntry({ ...validBase, ext: ".PNG" })).toBe(true);
+    });
+    test("mixed-case extensions accepted", () => {
+      expect(validateIndexEntry({ ...validBase, ext: ".PdF" })).toBe(true);
+    });
+    test("digits after dot accepted (e.g. .mp4, .mp3)", () => {
+      expect(validateIndexEntry({ ...validBase, ext: ".mp4" })).toBe(true);
+    });
+    test("exactly 16 chars after dot is accepted (upper bound of {1,16})", () => {
+      expect(validateIndexEntry({ ...validBase, ext: "." + "a".repeat(16) })).toBe(true);
+    });
+    test("exactly 17 chars after dot is rejected (immediately past upper bound)", () => {
+      expect(validateIndexEntry({ ...validBase, ext: "." + "a".repeat(17) })).toBe(false);
+    });
+    test("compound extensions like .tar.gz are rejected (contains inner dot)", () => {
+      // Documented: the current regex does not support compound
+      // extensions. If we ever want .tar.gz support, it needs its
+      // own explicit design + this row updated with lead sign-off.
+      expect(validateIndexEntry({ ...validBase, ext: ".tar.gz" })).toBe(false);
+    });
+  });
+});