linkshell-cli 0.5.2 → 0.5.5

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -4,7 +4,7 @@ import WebSocket from "ws";
4
4
  import { hostname, platform, homedir } from "node:os";
5
5
  import { writeFileSync, readFileSync, readdirSync, statSync, unlinkSync, mkdirSync, existsSync, openSync, readSync, closeSync } from "node:fs";
6
6
  import { tmpdir } from "node:os";
7
- import { join, basename, resolve } from "node:path";
7
+ import { join, basename, resolve, relative, isAbsolute } from "node:path";
8
8
  import {
9
9
  agentV2MessageRoute,
10
10
  createEnvelope,
@@ -21,7 +21,7 @@ import { ScreenShare } from "./screen-share.js";
21
21
  import { getLanIp } from "../utils/lan-ip.js";
22
22
  import { startKeepAwake, type KeepAwakeHandle } from "../utils/keep-awake.js";
23
23
  import { loadOrCreateMachineIdentity, type MachineIdentity } from "../machine-id.js";
24
- import { getValidToken } from "../auth.js";
24
+ import { getValidToken, refreshAccessToken } from "../auth.js";
25
25
  import { AgentSessionProxy } from "./acp/agent-session.js";
26
26
  import { AgentWorkspaceProxy } from "./acp/agent-workspace.js";
27
27
  import { detectAvailableProviders, type AgentProvider } from "./acp/provider-resolver.js";
@@ -43,14 +43,43 @@ export interface BridgeSessionOptions {
43
43
  agentUi?: boolean;
44
44
  agentProvider?: AgentProvider;
45
45
  agentCommand?: string;
46
+ /**
47
+ * Confinement root for client-driven filesystem operations
48
+ * (terminal.browse / terminal.file.read / terminal.mkdir).
49
+ * Defaults to the bridge's cwd. Paths resolving outside this root are rejected.
50
+ */
51
+ fileRoot?: string;
52
+ /**
53
+ * Ports the tunnel proxy is allowed to connect to on 127.0.0.1.
54
+ * Defaults to DEFAULT_TUNNEL_PORTS when omitted.
55
+ */
56
+ allowedTunnelPorts?: number[];
46
57
  }
47
58
 
48
59
  const HEARTBEAT_INTERVAL = 15_000;
60
+ // Transport-layer WS keepalive: ping the gateway periodically; if a pong
61
+ // doesn't arrive before the next ping, the socket is half-open (network
62
+ // silently dropped, no FIN) — terminate it so the reconnect loop kicks in.
63
+ const WS_PING_INTERVAL = 20_000;
49
64
  const RECONNECT_BASE_DELAY = 1_000;
50
65
  const RECONNECT_MAX_DELAY = 30_000;
51
- const RECONNECT_MAX_ATTEMPTS = 20;
66
+ // Cap the exponent so the backoff levels off at RECONNECT_MAX_DELAY instead of
67
+ // overflowing. There is NO attempt ceiling — a connected host never gives up
68
+ // (only `stop()`/all-PTYs-exited ends it), so it survives gateway restarts and
69
+ // arbitrarily long network outages, reconnecting whenever connectivity returns.
70
+ const RECONNECT_MAX_EXPONENT = 5;
52
71
  const DEFAULT_TERMINAL_ID = "default";
53
72
  const HOOK_BODY_LIMIT = 256 * 1024;
73
+ const SCROLLBACK_LINES = 500;
74
+ // Upper bound on concurrent live terminals — forceNew removed the cwd-dedup
75
+ // ceiling, so bound growth to avoid PTY/fd exhaustion from a runaway client.
76
+ const MAX_TERMINALS = 12;
77
+ // How long to keep an exited terminal (with its scrollback) in memory so the
78
+ // client can replay its final output before it is reaped.
79
+ const EXITED_TERMINAL_GRACE_MS = 30_000;
80
+ // Default ports the bridge will proxy tunnel requests to when no explicit
81
+ // allowlist is configured. Covers the common dev servers the app tunnels to.
82
+ const DEFAULT_TUNNEL_PORTS = [3000, 3001, 4321, 5173, 5174, 8080, 8000, 8081];
54
83
  const PERMISSION_REQUEST_TIMEOUT_MS = Number(
55
84
  process.env.LINKSHELL_PERMISSION_TIMEOUT_MS ?? 5 * 60_000,
56
85
  );
@@ -328,10 +357,15 @@ export class BridgeSession {
328
357
  private socket: WebSocket | undefined;
329
358
  private terminals = new Map<string, TerminalInstance>();
330
359
  private heartbeatTimer: ReturnType<typeof setInterval> | undefined;
360
+ private wsPingTimer: ReturnType<typeof setInterval> | undefined;
361
+ // True between sending a WS ping and receiving its pong. If still true at the
362
+ // next ping tick, the socket is half-open and gets terminated.
363
+ private awaitingPong = false;
331
364
  private reconnectTimer: ReturnType<typeof setTimeout> | undefined;
332
365
  private reconnectAttempts = 0;
333
366
  private reconnecting = false;
334
367
  private sessionId = "";
368
+ private hostToken: string | undefined;
335
369
  private exited = false;
336
370
  private stopped = false;
337
371
  private permissionStacks = new Map<string, Array<{
@@ -351,10 +385,51 @@ export class BridgeSession {
351
385
  private agentSession: AgentSessionProxy | undefined;
352
386
  private agentWorkspace: AgentWorkspaceProxy | undefined;
353
387
  private machineIdentity: MachineIdentity | undefined;
388
+ // Confinement root for client-driven filesystem ops; resolved once at construction.
389
+ private readonly fileRoot: string;
390
+ // Ports the tunnel proxy may connect to on 127.0.0.1.
391
+ private readonly allowedTunnelPorts: Set<number>;
392
+ // Pending deletions of exited terminals (terminalId → timer) so we can clear on stop().
393
+ private exitedTerminalTimers = new Map<string, ReturnType<typeof setTimeout>>();
394
+ // Invoked when all terminals have exited naturally — closes embedded gateway,
395
+ // removes daemon PID file, etc. Set by the CLI entry point.
396
+ private onAllTerminalsExited: (() => void | Promise<void>) | undefined;
354
397
 
355
398
  constructor(options: BridgeSessionOptions) {
356
399
  this.options = options;
357
400
  this.sessionId = options.sessionId ?? "";
401
+ // Confine client-driven file ops to the user's home by default (single-user
402
+ // self-hosted model: the owner browses their own projects under ~). This
403
+ // still blocks /etc, /usr, other users' dirs. Override via --file-root.
404
+ this.fileRoot = resolve(options.fileRoot ?? homedir());
405
+ this.allowedTunnelPorts = new Set(
406
+ options.allowedTunnelPorts && options.allowedTunnelPorts.length > 0
407
+ ? options.allowedTunnelPorts
408
+ : DEFAULT_TUNNEL_PORTS,
409
+ );
410
+ }
411
+
412
+ /** Register a callback invoked once all terminals exit naturally (process self-cleanup). */
413
+ setOnAllTerminalsExited(handler: () => void | Promise<void>): void {
414
+ this.onAllTerminalsExited = handler;
415
+ }
416
+
417
+ /**
418
+ * Resolve a client-supplied path inside the confinement root.
419
+ * Returns the absolute resolved path, or null if it escapes the root.
420
+ */
421
+ private resolveConfinedPath(clientPath: string): string | null {
422
+ const expanded = clientPath.startsWith("~")
423
+ ? clientPath.replace(/^~/, homedir())
424
+ : clientPath;
425
+ // Resolve relative paths against the confinement root, not process.cwd().
426
+ const target = isAbsolute(expanded)
427
+ ? resolve(expanded)
428
+ : resolve(this.fileRoot, expanded);
429
+ const rel = relative(this.fileRoot, target);
430
+ if (rel === "") return target; // the root itself
431
+ if (rel.startsWith("..") || isAbsolute(rel)) return null;
432
+ return target;
358
433
  }
359
434
 
360
435
  private log(msg: string): void {
@@ -373,6 +448,9 @@ export class BridgeSession {
373
448
  `starting session (gateway=${this.options.gatewayUrl}, provider=${this.options.providerConfig.provider})`,
374
449
  );
375
450
  this.machineIdentity = loadOrCreateMachineIdentity();
451
+ // Sweep stale LinkShell hook entries (dead curl ports) left by a prior crash
452
+ // before we write fresh ones for this session.
453
+ this.sweepStaleHookConfigs();
376
454
  if (!this.sessionId) {
377
455
  await this.createPairing();
378
456
  }
@@ -421,9 +499,14 @@ export class BridgeSession {
421
499
  const body = (await res.json()) as {
422
500
  sessionId: string;
423
501
  pairingCode: string;
502
+ hostToken?: string;
424
503
  expiresAt: string;
425
504
  };
426
505
  this.sessionId = body.sessionId;
506
+ // Secret token proving we are the legitimate host for this session; sent on
507
+ // every host WS connect so a third party who learns the sessionId can't
508
+ // hijack the host role.
509
+ this.hostToken = body.hostToken;
427
510
 
428
511
  const pairingGateway = resolvePairingGateway(
429
512
  this.options.gatewayHttpUrl,
@@ -474,17 +557,40 @@ export class BridgeSession {
474
557
  this.options.authToken = token;
475
558
  return token;
476
559
  }
560
+ // Refresh failed (e.g. offline / transient Supabase error). Do NOT clear
561
+ // the token — keep using the last one we had. It may be expired and the
562
+ // gateway may reject it, but the close handler force-refreshes on the next
563
+ // attempt, and clearing it would permanently downgrade us to no-auth and
564
+ // lock a pro host out of its own sessions once connectivity returns.
477
565
  process.stderr.write(
478
- "[bridge] login token expired and refresh failed; run `linkshell login` if the gateway rejects the connection\n",
566
+ "[bridge] token refresh failed; reusing last token (will retry refresh on reconnect)\n",
479
567
  );
480
- this.options.authToken = undefined;
481
- return undefined;
568
+ return this.options.authToken;
482
569
  } catch (error) {
483
570
  this.log(`failed to refresh login token: ${error instanceof Error ? error.message : String(error)}`);
484
571
  return this.options.authToken;
485
572
  }
486
573
  }
487
574
 
575
+ // After an auth-class WS close, force a token refresh before reconnecting so
576
+ // we don't spin on 401/4001/4003. A gateway restart surfaces as 1006 right
577
+ // after a fresh connect, so we refresh on that too (cheap, debounced by the
578
+ // reconnect backoff). No-op for clean/normal closes.
579
+ private async maybeRefreshTokenForClose(code: number): Promise<void> {
580
+ if (!this.options.authToken) return;
581
+ const authClass = code === 4001 || code === 4003 || code === 1006 || code === 401;
582
+ if (!authClass) return;
583
+ try {
584
+ const refreshed = await refreshAccessToken();
585
+ if (refreshed?.accessToken) {
586
+ this.options.authToken = refreshed.accessToken;
587
+ this.log("refreshed auth token after auth-class close");
588
+ }
589
+ } catch {
590
+ // Offline — keep the old token; the reconnect loop will retry later.
591
+ }
592
+ }
593
+
488
594
  private async connectGateway(): Promise<void> {
489
595
  if (this.stopped) {
490
596
  return;
@@ -494,11 +600,24 @@ export class BridgeSession {
494
600
  url.searchParams.set("sessionId", this.sessionId);
495
601
  url.searchParams.set("role", "host");
496
602
  const authToken = await this.resolveAuthToken();
603
+ const wsOptions: WebSocket.ClientOptions = {};
604
+ const headers: Record<string, string> = {};
497
605
  if (authToken) {
606
+ // Prefer the Authorization header so the token doesn't leak into URLs/logs.
607
+ headers["Authorization"] = `Bearer ${authToken}`;
608
+ // TODO: remove the query param once all gateways read the Authorization
609
+ // header. Kept for backward compatibility with older gateways that only
610
+ // read the `auth_token` query parameter.
498
611
  url.searchParams.set("auth_token", authToken);
499
612
  }
613
+ if (this.hostToken) {
614
+ headers["x-linkshell-host-token"] = this.hostToken;
615
+ }
616
+ if (Object.keys(headers).length > 0) {
617
+ wsOptions.headers = headers;
618
+ }
500
619
 
501
- this.socket = new WebSocket(url);
620
+ this.socket = new WebSocket(url, wsOptions);
502
621
 
503
622
  this.socket.on("open", () => {
504
623
  process.stderr.write(
@@ -526,6 +645,12 @@ export class BridgeSession {
526
645
  }),
527
646
  );
528
647
  this.startHeartbeat();
648
+ this.startWsPing();
649
+ });
650
+
651
+ // Gateway answered our WS ping — connection is alive.
652
+ this.socket.on("pong", () => {
653
+ this.awaitingPong = false;
529
654
  });
530
655
 
531
656
  this.socket.on("message", (data) => {
@@ -546,18 +671,32 @@ export class BridgeSession {
546
671
 
547
672
  this.socket.on("close", (code, reasonBuffer) => {
548
673
  this.stopHeartbeat();
674
+ this.stopWsPing();
549
675
  this.socket = undefined;
550
676
  const reason = reasonBuffer.toString();
551
- process.stderr.write(
552
- `[bridge] gateway connection closed (code=${code}${reason ? `, reason=${reason}` : ""})\n`,
553
- );
677
+ // Quiet during an ongoing outage: log the first close (we were connected,
678
+ // attempts==0) and every 10th, so multi-hour outages don't flood the log.
679
+ if (this.reconnectAttempts === 0 || this.reconnectAttempts % 10 === 0) {
680
+ process.stderr.write(
681
+ `[bridge] gateway connection closed (code=${code}${reason ? `, reason=${reason}` : ""})\n`,
682
+ );
683
+ }
554
684
  if (!this.exited) {
555
- this.scheduleReconnect();
685
+ // Auth-class closes (token expired / gateway restarted) get a forced
686
+ // token refresh before the next attempt, so we don't loop on 401/4xxx.
687
+ this.maybeRefreshTokenForClose(code).finally(() => {
688
+ this.scheduleReconnect();
689
+ });
556
690
  }
557
691
  });
558
692
 
559
693
  this.socket.on("error", (error) => {
560
- process.stderr.write(`[bridge] gateway error: ${error.message}\n`);
694
+ // During an outage every retry emits a connection error; the `close`
695
+ // handler already logs (quietly) and drives reconnect. Only surface the
696
+ // first error of an outage to avoid flooding the daemon log.
697
+ if (this.reconnectAttempts <= 1) {
698
+ process.stderr.write(`[bridge] gateway error: ${error.message || "connection failed"}\n`);
699
+ }
561
700
  });
562
701
  }
563
702
 
@@ -625,10 +764,24 @@ export class BridgeSession {
625
764
  case "terminal.spawn": {
626
765
  const p = parseTypedPayload("terminal.spawn", envelope.payload);
627
766
  const normalizedCwd = resolve(p.cwd);
628
- // Dedup: if a running terminal already exists for this cwd, return it
629
- const existing = [...this.terminals.values()].find(
630
- (t) => t.status === "running" && resolve(t.cwd) === normalizedCwd,
631
- );
767
+ // Cap concurrent terminals — forceNew removed the cwd-dedup ceiling, so
768
+ // bound growth explicitly to avoid PTY/fd exhaustion from a runaway client.
769
+ const liveCount = [...this.terminals.values()].filter((t) => t.status === "running").length;
770
+ if (liveCount >= MAX_TERMINALS) {
771
+ this.send(createEnvelope({
772
+ type: "session.error",
773
+ sessionId: this.sessionId,
774
+ payload: { code: "too_many_terminals", message: `最多同时打开 ${MAX_TERMINALS} 个终端` },
775
+ }));
776
+ break;
777
+ }
778
+ // Dedup by cwd UNLESS the client explicitly asks for a fresh terminal
779
+ // (forceNew lets web open multiple tabs in the same directory).
780
+ const existing = p.forceNew
781
+ ? undefined
782
+ : [...this.terminals.values()].find(
783
+ (t) => t.status === "running" && resolve(t.cwd) === normalizedCwd,
784
+ );
632
785
  if (existing) {
633
786
  this.send(createEnvelope({
634
787
  type: "terminal.spawned",
@@ -669,9 +822,15 @@ export class BridgeSession {
669
822
  }
670
823
  case "terminal.browse": {
671
824
  const p = parseTypedPayload("terminal.browse", envelope.payload);
672
- // Expand ~ to home directory
673
- const rawPath = p.path.startsWith("~") ? p.path.replace(/^~/, homedir()) : p.path;
674
- const browsePath = resolve(rawPath);
825
+ const browsePath = this.resolveConfinedPath(p.path);
826
+ if (browsePath === null) {
827
+ this.send(createEnvelope({
828
+ type: "terminal.browse.result",
829
+ sessionId: this.sessionId,
830
+ payload: { path: p.path, entries: [], error: "Path is outside the allowed root", requestId: p.requestId },
831
+ }));
832
+ break;
833
+ }
675
834
  try {
676
835
  const entries = readdirSync(browsePath, { withFileTypes: true })
677
836
  .filter((d) => !d.name.startsWith(".") && (d.isDirectory() || (p.includeFiles && d.isFile())))
@@ -706,8 +865,22 @@ export class BridgeSession {
706
865
  }
707
866
  case "terminal.file.read": {
708
867
  const p = parseTypedPayload("terminal.file.read", envelope.payload);
709
- const rawPath = p.path.startsWith("~") ? p.path.replace(/^~/, homedir()) : p.path;
710
- const filePath = resolve(rawPath);
868
+ const filePath = this.resolveConfinedPath(p.path);
869
+ if (filePath === null) {
870
+ this.send(createEnvelope({
871
+ type: "terminal.file.read.result",
872
+ sessionId: this.sessionId,
873
+ payload: {
874
+ path: p.path,
875
+ content: "",
876
+ encoding: "utf8",
877
+ truncated: false,
878
+ error: "Path is outside the allowed root",
879
+ requestId: p.requestId,
880
+ },
881
+ }));
882
+ break;
883
+ }
711
884
  try {
712
885
  const stats = statSync(filePath);
713
886
  if (!stats.isFile()) {
@@ -755,8 +928,15 @@ export class BridgeSession {
755
928
  }
756
929
  case "terminal.mkdir": {
757
930
  const p = parseTypedPayload("terminal.mkdir", envelope.payload);
758
- const rawPath = p.path.startsWith("~") ? p.path.replace(/^~/, homedir()) : p.path;
759
- const dirPath = resolve(rawPath);
931
+ const dirPath = this.resolveConfinedPath(p.path);
932
+ if (dirPath === null) {
933
+ this.send(createEnvelope({
934
+ type: "terminal.browse.result",
935
+ sessionId: this.sessionId,
936
+ payload: { path: p.path, entries: [], error: "Path is outside the allowed root" },
937
+ }));
938
+ break;
939
+ }
760
940
  try {
761
941
  mkdirSync(dirPath, { recursive: true });
762
942
  // Browse the parent to refresh the listing
@@ -1017,6 +1197,21 @@ export class BridgeSession {
1017
1197
  }): void {
1018
1198
  const { requestId, method, url: reqUrl, headers, body, port } = payload;
1019
1199
 
1200
+ // SSRF guard: only proxy to ports in the allowlist (pinned to 127.0.0.1).
1201
+ if (!this.allowedTunnelPorts.has(port)) {
1202
+ this.log(`rejecting tunnel request to disallowed port ${port}`);
1203
+ if (headers.upgrade === "websocket") {
1204
+ this.send(createEnvelope({
1205
+ type: "tunnel.ws.close",
1206
+ sessionId: this.sessionId,
1207
+ payload: { requestId, code: 1008, reason: `Port ${port} is not allowed` },
1208
+ }));
1209
+ } else {
1210
+ this.sendTunnelError(requestId, 403, `Port ${port} is not allowed`);
1211
+ }
1212
+ return;
1213
+ }
1214
+
1020
1215
  // WebSocket upgrade request
1021
1216
  if (headers.upgrade === "websocket") {
1022
1217
  this.handleTunnelWsUpgrade(requestId, port, reqUrl);
@@ -1278,7 +1473,7 @@ export class BridgeSession {
1278
1473
  cwd,
1279
1474
  projectName: basename(cwd),
1280
1475
  provider,
1281
- scrollback: new ScrollbackBuffer(1000),
1476
+ scrollback: new ScrollbackBuffer(SCROLLBACK_LINES),
1282
1477
  outputSeq: 0,
1283
1478
  statusSeq: 0,
1284
1479
  status: "running",
@@ -1318,6 +1513,20 @@ export class BridgeSession {
1318
1513
  }));
1319
1514
  this.sendTerminalList();
1320
1515
 
1516
+ // Reap the exited terminal (and its scrollback) after a grace period so
1517
+ // clients can replay final output, then free the memory. Tracked so the
1518
+ // timer can be cleared on stop().
1519
+ const reapTimer = setTimeout(() => {
1520
+ this.exitedTerminalTimers.delete(terminalId);
1521
+ const t = this.terminals.get(terminalId);
1522
+ if (t && t.status === "exited") {
1523
+ this.terminals.delete(terminalId);
1524
+ this.log(`reaped exited terminal ${terminalId}`);
1525
+ }
1526
+ }, EXITED_TERMINAL_GRACE_MS);
1527
+ if (typeof reapTimer.unref === "function") reapTimer.unref();
1528
+ this.exitedTerminalTimers.set(terminalId, reapTimer);
1529
+
1321
1530
  // If all terminals exited, close the session
1322
1531
  const allExited = [...this.terminals.values()].every((t) => t.status === "exited");
1323
1532
  if (allExited) {
@@ -1325,6 +1534,9 @@ export class BridgeSession {
1325
1534
  setTimeout(() => {
1326
1535
  this.stopHeartbeat();
1327
1536
  this.socket?.close();
1537
+ // Self-clean process-level resources (embedded gateway + PID file)
1538
+ // since the PTYs exited naturally rather than via a signal handler.
1539
+ void this.onAllTerminalsExited?.();
1328
1540
  }, 500);
1329
1541
  process.exitCode = exitCode ?? 0;
1330
1542
  }
@@ -2046,6 +2258,59 @@ export class BridgeSession {
2046
2258
  } catch { /* ignore parse errors */ }
2047
2259
  }
2048
2260
 
2261
+ /**
2262
+ * Remove ALL LinkShell hook entries (any marker) from a JSON config file.
2263
+ * Used on startup to sweep stale entries left behind by a crashed instance,
2264
+ * whose curl ports are now dead. Matches via isLinkShellHookEntry rather than
2265
+ * a single marker so entries from previous runs are also cleaned.
2266
+ */
2267
+ private sweepLinkShellHookEntries(configPath: string): void {
2268
+ if (!existsSync(configPath)) return;
2269
+ try {
2270
+ // Re-read immediately before write to minimize the window for a
2271
+ // concurrent writer to clobber our changes.
2272
+ const raw = JSON.parse(readFileSync(configPath, "utf8"));
2273
+ const hooks = raw.hooks as Record<string, unknown[]> | undefined;
2274
+ if (!hooks) return;
2275
+
2276
+ let changed = false;
2277
+ for (const [eventName, entries] of Object.entries(hooks)) {
2278
+ if (!Array.isArray(entries)) continue;
2279
+ const filtered = entries.filter((entry) => !isLinkShellHookEntry(entry));
2280
+ if (filtered.length !== entries.length) {
2281
+ changed = true;
2282
+ if (filtered.length === 0) {
2283
+ delete hooks[eventName];
2284
+ } else {
2285
+ hooks[eventName] = filtered;
2286
+ }
2287
+ }
2288
+ }
2289
+
2290
+ if (changed) {
2291
+ if (Object.keys(hooks).length === 0) {
2292
+ delete raw.hooks;
2293
+ }
2294
+ writeFileSync(configPath, JSON.stringify(raw, null, 2));
2295
+ this.log(`swept stale LinkShell hook entries from ${configPath}`);
2296
+ }
2297
+ } catch { /* ignore parse errors */ }
2298
+ }
2299
+
2300
+ /** Sweep stale LinkShell hook entries from known shared config files on startup. */
2301
+ private sweepStaleHookConfigs(): void {
2302
+ const home = homedir();
2303
+ const candidates = [
2304
+ join(home, ".claude", "settings.json"),
2305
+ join(home, ".codex", "hooks.json"),
2306
+ join(home, ".gemini", "settings.json"),
2307
+ join(process.cwd(), "hooks.json"), // copilot (per-cwd)
2308
+ ];
2309
+ for (const path of candidates) {
2310
+ this.sweepLinkShellHookEntries(path);
2311
+ }
2312
+ }
2313
+
2049
2314
  private send(message: Envelope): void {
2050
2315
  if (!this.socket || this.socket.readyState !== WebSocket.OPEN) {
2051
2316
  return;
@@ -2089,6 +2354,45 @@ export class BridgeSession {
2089
2354
  }
2090
2355
  }
2091
2356
 
2357
+ // Transport-layer keepalive. WS ping/pong detects a half-open socket (network
2358
+ // dropped without a TCP FIN, e.g. laptop sleep / Wi-Fi loss) that would
2359
+ // otherwise hang silently. If the previous ping's pong never arrived, the
2360
+ // socket is dead → terminate it, which fires `close` → scheduleReconnect.
2361
+ private startWsPing(): void {
2362
+ this.stopWsPing();
2363
+ this.awaitingPong = false;
2364
+ this.wsPingTimer = setInterval(() => {
2365
+ const sock = this.socket;
2366
+ if (!sock || sock.readyState !== WebSocket.OPEN) return;
2367
+ if (this.awaitingPong) {
2368
+ // No pong since the last ping → half-open. Forcibly close so we reconnect.
2369
+ process.stderr.write("[bridge] no pong from gateway; terminating dead connection\n");
2370
+ this.awaitingPong = false;
2371
+ try {
2372
+ sock.terminate();
2373
+ } catch {
2374
+ // terminate may throw if already closing; close handler still runs.
2375
+ }
2376
+ return;
2377
+ }
2378
+ this.awaitingPong = true;
2379
+ try {
2380
+ sock.ping();
2381
+ } catch {
2382
+ // ping failed → let the next tick terminate, or close already fired.
2383
+ }
2384
+ }, WS_PING_INTERVAL);
2385
+ if (typeof this.wsPingTimer.unref === "function") this.wsPingTimer.unref();
2386
+ }
2387
+
2388
+ private stopWsPing(): void {
2389
+ if (this.wsPingTimer) {
2390
+ clearInterval(this.wsPingTimer);
2391
+ this.wsPingTimer = undefined;
2392
+ }
2393
+ this.awaitingPong = false;
2394
+ }
2395
+
2092
2396
  private startScreenCapture(fps: number, quality: number, scale: number): void {
2093
2397
  if (!this.options.screen) {
2094
2398
  this.log("screen sharing not enabled (use --screen)");
@@ -2152,25 +2456,24 @@ export class BridgeSession {
2152
2456
  }
2153
2457
 
2154
2458
  private scheduleReconnect(): void {
2155
- if (this.reconnecting) return;
2459
+ if (this.reconnecting || this.stopped || this.exited) return;
2156
2460
 
2157
- // In daemon mode, never give up — reset attempts after hitting max
2158
- if (this.reconnectAttempts >= RECONNECT_MAX_ATTEMPTS) {
2461
+ this.reconnecting = true;
2462
+ // Exponential backoff capped at RECONNECT_MAX_DELAY, with full jitter so a
2463
+ // fleet of hosts doesn't stampede a restarting gateway. The attempt counter
2464
+ // only ever grows toward the cap — it is NEVER reset to "give up"; the host
2465
+ // keeps trying forever until stop()/exit.
2466
+ const exponent = Math.min(this.reconnectAttempts, RECONNECT_MAX_EXPONENT);
2467
+ const base = Math.min(RECONNECT_BASE_DELAY * 2 ** exponent, RECONNECT_MAX_DELAY);
2468
+ const delay = Math.round(base * (0.5 + Math.random() * 0.5));
2469
+ this.reconnectAttempts++;
2470
+ // Quiet logging: only the first retry of an outage and every 10th attempt,
2471
+ // so a multi-hour offline window doesn't flood the daemon log.
2472
+ if (this.reconnectAttempts === 1 || this.reconnectAttempts % 10 === 0) {
2159
2473
  process.stderr.write(
2160
- "[bridge] max reconnect attempts reached, resetting counter and continuing...\n",
2474
+ `[bridge] reconnecting in ${delay}ms (attempt ${this.reconnectAttempts})\n`,
2161
2475
  );
2162
- this.reconnectAttempts = 0;
2163
2476
  }
2164
-
2165
- this.reconnecting = true;
2166
- const delay = Math.min(
2167
- RECONNECT_BASE_DELAY * 2 ** this.reconnectAttempts,
2168
- RECONNECT_MAX_DELAY,
2169
- );
2170
- this.reconnectAttempts++;
2171
- process.stderr.write(
2172
- `[bridge] reconnecting in ${delay}ms (attempt ${this.reconnectAttempts})\n`,
2173
- );
2174
2477
  this.reconnectTimer = setTimeout(() => {
2175
2478
  this.reconnectTimer = undefined;
2176
2479
  if (this.stopped || this.exited) {
@@ -2189,6 +2492,7 @@ export class BridgeSession {
2189
2492
  this.stopped = true;
2190
2493
  this.exited = true;
2191
2494
  this.stopHeartbeat();
2495
+ this.stopWsPing();
2192
2496
  this.stopScreenCapture();
2193
2497
  this.agentSession?.stop();
2194
2498
  this.agentSession = undefined;
@@ -2207,6 +2511,11 @@ export class BridgeSession {
2207
2511
  ws.close(1001, "Session stopped");
2208
2512
  }
2209
2513
  this.tunnelSockets.clear();
2514
+ // Clear pending exited-terminal reap timers.
2515
+ for (const timer of this.exitedTerminalTimers.values()) {
2516
+ clearTimeout(timer);
2517
+ }
2518
+ this.exitedTerminalTimers.clear();
2210
2519
  for (const term of this.terminals.values()) {
2211
2520
  this.cleanupHookServer(term);
2212
2521
  if (term.status === "running") term.pty.kill();