fix(ios-qa): reuse rotated bearer so /ios-qa drives a real device across sessions

bin/gstack-ios-qa-daemon

@@ -13,6 +13,12 @@
 #                                  the first paired connected device is used)
 #   GSTACK_IOS_TARGET_BUNDLE_ID  — bundle ID of the iOS app hosting StateServer
 #                                  (default com.gstack.iosqa.fixture)
+#   GSTACK_IOS_LAUNCH_ENV        — JSON dict of env vars set when the daemon
+#                                  cold-launches the app, forwarded to
+#                                  `devicectl ... --environment-variables`.
+#                                  Required for apps that gate their debug
+#                                  bridge behind a flag, e.g. BuckHound:
+#                                  '{"BH_ENABLE_IOS_QA_BRIDGE":"1"}'
 #
 # Readiness protocol: prints `READY: port=<n> pid=<pid>` to stdout once both
 # listeners are bound. Spawners read stdin with a ~5s timeout to confirm.

ios-qa/daemon/src/devicectl.ts

@@ -270,8 +270,17 @@ export function launchApp(
   udid: string,
   bundleId: string,
   spawn: SpawnImpl = defaultSpawn,
+  env?: Record<string, string>,
 ): { ok: boolean; error?: string } {
-  const r = spawn('xcrun', ['devicectl', 'device', 'process', 'launch', '--device', udid, bundleId]);
+  const args = ['devicectl', 'device', 'process', 'launch', '--device', udid];
+  // Forward launch-time env vars (e.g. an app's debug-bridge enable flag) so a
+  // cold start actually boots the in-app StateServer. Bundle id stays the
+  // trailing positional arg, after every flag.
+  if (env && Object.keys(env).length > 0) {
+    args.push('--environment-variables', JSON.stringify(env));
+  }
+  args.push(bundleId);
+  const r = spawn('xcrun', args);
   if (r.status === 0) return { ok: true };
   const err = (r.stderr?.toString() ?? '') + (r.stdout?.toString() ?? '');
   if (err.includes('was not, or could not be, unlocked')) {

ios-qa/daemon/src/index.ts

@@ -20,7 +20,7 @@ import { SessionTokenStore } from './session-tokens';
 import { mintForCaller } from './auth-mint';
 import { classifyRoute, proxyToDevice, type DeviceTunnel } from './proxy';
 import { writeAudit, writeAttempt, sanitizeReplacer } from './audit';
-import { bootstrapTunnel } from './tunnel-bootstrap';
+import { acquireTunnel } from './session-cache';
 import { startTunnelKeepalive } from './devicectl';
 import type { Capability } from './types';
 
@@ -399,6 +399,18 @@ if (import.meta.main) {
   const tailnet = process.argv.includes('--tailnet');
   const targetUDID = process.env.GSTACK_IOS_TARGET_UDID;
   const bundleId = process.env.GSTACK_IOS_TARGET_BUNDLE_ID ?? 'com.gstack.iosqa.fixture';
+  // Env vars to set when the daemon cold-launches the app. Apps that gate their
+  // debug bridge behind a flag (BuckHound: BH_ENABLE_IOS_QA_BRIDGE=1) MUST set
+  // this, or a cold start launches the app without the bridge and the
+  // StateServer never binds. Malformed JSON is ignored (warn + no launch env).
+  let launchEnv: Record<string, string> | undefined;
+  if (process.env.GSTACK_IOS_LAUNCH_ENV) {
+    try {
+      launchEnv = JSON.parse(process.env.GSTACK_IOS_LAUNCH_ENV) as Record<string, string>;
+    } catch {
+      process.stderr.write('GSTACK_IOS_LAUNCH_ENV is not valid JSON; ignoring\n');
+    }
+  }
 
   // Default tunnelProvider: when GSTACK_IOS_TARGET_UDID (or a default with
   // any connected paired device) is set, bootstrap a real CoreDevice tunnel.
@@ -410,17 +422,22 @@ if (import.meta.main) {
   // without a poke every few seconds the IPv6 becomes unroutable.
   let keepalive: { stop: () => void } | null = null;
   const realTunnelProvider = async () => {
-    const result = await bootstrapTunnel({
+    // acquireTunnel reuses a cached rotated bearer (warm-start) when the device
+    // still honors it, and only falls back to a full single-use boot-token
+    // rotate on a genuinely fresh app launch. This is what lets a real device
+    // be driven across tunnel-cache refreshes, daemon restarts, and repeat
+    // /ios-qa sessions instead of exactly once per app launch.
+    const tunnel = await acquireTunnel({
       udid: targetUDID,
       bundleId,
+      port: 9999, // in-app StateServer port (not the daemon's loopback port)
+      launchEnv,
+      logImpl: (m) => process.stderr.write(m + '\n'),
     });
-    if (!result.ok) {
-      process.stderr.write(`bootstrap error: ${result.error}${result.detail ? ' — ' + result.detail : ''}\n`);
-      return null;
-    }
+    if (!tunnel) return null;
     if (keepalive) keepalive.stop();
-    keepalive = startTunnelKeepalive(result.tunnel.udid);
-    return result.tunnel;
+    keepalive = startTunnelKeepalive(tunnel.udid);
+    return tunnel;
   };
 
   const shutdown = () => {

ios-qa/daemon/src/session-cache.ts

@@ -0,0 +1,176 @@
+// Rotated-token session cache + reuse. This is the Mac-side half of the
+// "bootstrap once per app launch" contract that SKILL.md Phase 0 (warm-start)
+// describes but the daemon never implemented.
+//
+// WHY this exists: the in-app StateServer boot token is single-use. The first
+// POST /auth/rotate sets bootTokenValid=false AND deletes the on-disk token
+// file. The daemon, however, re-runs the full bootstrap (copy boot token +
+// rotate) on every tunnel-cache refresh, daemon restart, and new /ios-qa
+// session. The second bootstrap then finds the token file gone ->
+// `boot_token_unavailable`, so a real device can be driven exactly once per
+// app launch. Persisting the rotated bearer the daemon already holds, and
+// reusing it after a cheap authenticated probe, makes re-bootstrap unnecessary.
+//
+// SECURITY: this changes nothing on the device. The StateServer stays
+// loopback-only, every endpoint still requires the rotated bearer, and the boot
+// token stays single-use. The only new artifact is a Mac-side 0600 cache file
+// holding the rotated bearer (the same value SKILL.md Phase 0 already specifies
+// the session cache holds), valid only while the app stays launched — a probe
+// that returns 401 means the app was relaunched, so we drop the stale token and
+// re-bootstrap from a freshly written boot token.
+
+import { readFileSync, writeFileSync, mkdirSync, chmodSync, rmSync } from 'fs';
+import { homedir } from 'os';
+import { dirname, join } from 'path';
+import { bootstrapTunnel, type BootstrapOptions } from './tunnel-bootstrap';
+import { getDeviceTunnelIPv6FromDevicectl, type SpawnImpl, type ResolveImpl } from './devicectl';
+import type { DeviceTunnel } from './proxy';
+
+export interface SessionCache {
+  udid: string;
+  bundleId: string;
+  port: number;
+  rotatedToken: string;
+  ipv6: string;
+  createdAt: number;
+}
+
+export function defaultSessionCachePath(): string {
+  return process.env.GSTACK_IOS_SESSION_CACHE
+    ?? join(homedir(), '.gstack', 'ios-qa-session.json');
+}
+
+/** Read the session cache. Returns null on missing/unreadable/corrupt file. */
+export function readSessionCache(path: string = defaultSessionCachePath()): SessionCache | null {
+  try {
+    const obj = JSON.parse(readFileSync(path, 'utf-8')) as Partial<SessionCache>;
+    if (
+      obj && typeof obj.udid === 'string' && typeof obj.bundleId === 'string'
+      && typeof obj.port === 'number' && typeof obj.rotatedToken === 'string'
+      && typeof obj.ipv6 === 'string' && typeof obj.createdAt === 'number'
+    ) {
+      return obj as SessionCache;
+    }
+    return null;
+  } catch {
+    return null;
+  }
+}
+
+/** Persist the session cache with owner-only (0600) perms. */
+export function writeSessionCache(cache: SessionCache, path: string = defaultSessionCachePath()): void {
+  mkdirSync(dirname(path), { recursive: true });
+  writeFileSync(path, JSON.stringify(cache), { mode: 0o600 });
+  // writeFileSync's mode only applies on create; force 0600 on overwrite too.
+  chmodSync(path, 0o600);
+}
+
+/** Remove the session cache (best-effort). */
+export function clearSessionCache(path: string = defaultSessionCachePath()): void {
+  try { rmSync(path, { force: true }); } catch { /* ignore */ }
+}
+
+export interface AcquireTunnelOptions {
+  udid?: string;
+  bundleId: string;
+  port: number;
+  /** Env vars to set if a cold-start bootstrap has to launch the app. */
+  launchEnv?: Record<string, string>;
+  // Injection seams (real defaults wire the production impls).
+  cachePath?: string;
+  readCacheImpl?: (path?: string) => SessionCache | null;
+  writeCacheImpl?: (cache: SessionCache, path?: string) => void;
+  clearCacheImpl?: (path?: string) => void;
+  resolveIPv6Impl?: (udid: string) => string | null | Promise<string | null>;
+  probeImpl?: (ipv6: string, port: number, token: string) => Promise<number>;
+  bootstrapImpl?: (opts: BootstrapOptions) => Promise<import('./tunnel-bootstrap').BootstrapResult>;
+  /** Diagnostic sink (defaults to no-op; the CLI wires it to stderr). */
+  logImpl?: (msg: string) => void;
+  // Pass-throughs to the underlying bootstrap.
+  spawnImpl?: SpawnImpl;
+  resolveImpl?: ResolveImpl;
+  fetchImpl?: typeof fetch;
+}
+
+/**
+ * Acquire a usable DeviceTunnel, reusing a cached rotated bearer when the
+ * device still honors it and only falling back to a full boot-token bootstrap
+ * when there is no usable cache or the cached bearer is rejected (app
+ * relaunched). Returns null on a transient device-unreachable condition (cache
+ * preserved) or a failed bootstrap.
+ */
+export async function acquireTunnel(opts: AcquireTunnelOptions): Promise<DeviceTunnel | null> {
+  const cachePath = opts.cachePath;
+  const readCache = opts.readCacheImpl ?? readSessionCache;
+  const writeCache = opts.writeCacheImpl ?? writeSessionCache;
+  const clearCache = opts.clearCacheImpl ?? clearSessionCache;
+  const resolveIPv6 = opts.resolveIPv6Impl ?? ((udid: string) => getDeviceTunnelIPv6FromDevicectl(udid, opts.spawnImpl));
+  const probe = opts.probeImpl ?? defaultProbe(opts.fetchImpl);
+  const bootstrap = opts.bootstrapImpl ?? bootstrapTunnel;
+  const log = opts.logImpl ?? (() => {});
+
+  const cache = readCache(cachePath);
+  const cacheUsable = !!cache
+    && cache.bundleId === opts.bundleId
+    && cache.port === opts.port
+    && (!opts.udid || cache.udid === opts.udid);
+
+  if (cache && cacheUsable) {
+    const ipv6 = await resolveIPv6(cache.udid);
+    if (!ipv6) { log('device unresolvable; keeping cached session token, will retry'); return null; }
+    const status = await probe(ipv6, cache.port, cache.rotatedToken);
+    if (status === 200) {
+      if (ipv6 !== cache.ipv6) writeCache({ ...cache, ipv6 }, cachePath);
+      log(`reusing cached session token for ${cache.udid} (no app relaunch needed)`);
+      return { udid: cache.udid, ipv6Addr: ipv6, port: cache.port, bootTokenRotated: cache.rotatedToken };
+    }
+    if (status === 401 || status === 403) {
+      log(`cached session token rejected (HTTP ${status}); app was relaunched — re-bootstrapping`);
+      clearCache(cachePath); // app relaunched -> stale rotated token; re-bootstrap below
+    } else {
+      log(`device unreachable during token probe (HTTP ${status}); keeping cached token, will retry`);
+      return null; // 0 (connection error) / 5xx — transient; do not discard a good token
+    }
+  }
+
+  const result = await bootstrap({
+    udid: opts.udid,
+    bundleId: opts.bundleId,
+    port: opts.port,
+    launchEnv: opts.launchEnv,
+    spawnImpl: opts.spawnImpl,
+    resolveImpl: opts.resolveImpl,
+    fetchImpl: opts.fetchImpl,
+  });
+  if (!result.ok) {
+    log(`bootstrap error: ${result.error}${result.detail ? ' — ' + result.detail : ''}`);
+    return null;
+  }
+  log(`bootstrapped fresh session token for ${result.tunnel.udid}`);
+
+  writeCache({
+    udid: result.tunnel.udid,
+    bundleId: opts.bundleId,
+    port: opts.port,
+    rotatedToken: result.tunnel.bootTokenRotated,
+    ipv6: result.tunnel.ipv6Addr,
+    createdAt: Date.now(),
+  }, cachePath);
+  return result.tunnel;
+}
+
+function defaultProbe(fetchFn: typeof fetch = fetch) {
+  return async (ipv6: string, port: number, token: string): Promise<number> => {
+    const isIPv6 = (ipv6.match(/:/g)?.length ?? 0) >= 2;
+    const host = isIPv6 ? `[${ipv6}]` : ipv6;
+    try {
+      const r = await fetchFn(`http://${host}:${port}/state/snapshot`, {
+        headers: { Authorization: `Bearer ${token}` },
+        signal: AbortSignal.timeout(4_000),
+      });
+      return r.status;
+    } catch {
+      return 0; // connection refused / timeout / tunnel down
+    }
+  };
+}

ios-qa/daemon/src/tunnel-bootstrap.ts

@@ -34,6 +34,11 @@ export interface BootstrapOptions {
   port?: number;
   /** Token-path inside the app sandbox (relative to data container). */
   bootTokenPath?: string;
+  /** Env vars to set when the daemon launches the app (cold start). Forwarded
+   *  to `devicectl ... --environment-variables`. Apps that gate their debug
+   *  bridge behind a flag (e.g. BuckHound's BH_ENABLE_IOS_QA_BRIDGE) need this
+   *  or a cold launch never boots the StateServer. */
+  launchEnv?: Record<string, string>;
   /** Max time to wait for the StateServer to start after launch (ms). */
   startupTimeoutMs?: number;
   /** Test injection. */
@@ -91,7 +96,7 @@ export async function bootstrapTunnel(opts: BootstrapOptions): Promise<Bootstrap
 
   // Step 2: launch app (idempotent — devicectl returns success if already running)
   if (!isAppRunning(target.identifier, opts.bundleId, spawn)) {
-    const launched = launchApp(target.identifier, opts.bundleId, spawn);
+    const launched = launchApp(target.identifier, opts.bundleId, spawn, opts.launchEnv);
     if (!launched.ok) {
       return { ok: false, error: launched.error === 'device_locked' ? 'device_locked' : 'launch_failed', detail: launched.error };
     }

ios-qa/daemon/test/launch-env.test.ts

@@ -0,0 +1,95 @@
+// Cold-start launch env. BuckHound (and any app that gates its debug bridge
+// behind an env var) needs the daemon to pass that var when it launches the
+// app itself. Without it, a cold start (app not already running) launches the
+// app WITHOUT the bridge -> the StateServer never binds -> state_server_unreachable,
+// and `/ios-qa` only works if the operator pre-launches by hand. The daemon
+// reads GSTACK_IOS_LAUNCH_ENV (a JSON dict) and forwards it to
+// `devicectl device process launch --environment-variables`.
+
+import { describe, test, expect } from 'bun:test';
+import { launchApp, type SpawnImpl } from '../src/devicectl';
+import { bootstrapTunnel } from '../src/tunnel-bootstrap';
+import { writeFileSync } from 'fs';
+
+function makeReturn(exit: number, stdout = '', stderr = '') {
+  return {
+    pid: 0,
+    output: [null, Buffer.from(stdout), Buffer.from(stderr)],
+    stdout: Buffer.from(stdout),
+    stderr: Buffer.from(stderr),
+    status: exit,
+    signal: null,
+  } as ReturnType<SpawnImpl>;
+}
+
+describe('launchApp environment variables', () => {
+  test('passes --environment-variables with the JSON dict, bundle id stays last', () => {
+    let captured: string[] = [];
+    const spawn: SpawnImpl = ((_cmd: string, args: string[]) => { captured = args; return makeReturn(0); }) as SpawnImpl;
+    const r = launchApp('UDID-1', 'com.test.app', spawn, { BH_ENABLE_IOS_QA_BRIDGE: '1' });
+    expect(r.ok).toBe(true);
+    const i = captured.indexOf('--environment-variables');
+    expect(i).toBeGreaterThan(-1);
+    expect(JSON.parse(captured[i + 1]!)).toEqual({ BH_ENABLE_IOS_QA_BRIDGE: '1' });
+    expect(captured[captured.length - 1]).toBe('com.test.app'); // bundle id is the trailing positional
+  });
+
+  test('omits --environment-variables when env is undefined or empty', () => {
+    for (const env of [undefined, {}]) {
+      let captured: string[] = [];
+      const spawn: SpawnImpl = ((_cmd: string, args: string[]) => { captured = args; return makeReturn(0); }) as SpawnImpl;
+      launchApp('UDID-1', 'com.test.app', spawn, env);
+      expect(captured.includes('--environment-variables')).toBe(false);
+      expect(captured[captured.length - 1]).toBe('com.test.app');
+    }
+  });
+});
+
+describe('bootstrapTunnel forwards launchEnv on cold start', () => {
+  test('threads launchEnv into the launch when the app is not already running', async () => {
+    const calls: string[][] = [];
+    const spawn: SpawnImpl = ((_cmd: string, args: string[]) => {
+      calls.push(args);
+      const joined = args.join(' ');
+      const writeJson = (obj: object) => {
+        const fi = args.indexOf('--json-output');
+        if (fi !== -1 && args[fi + 1]) writeFileSync(args[fi + 1]!, JSON.stringify(obj));
+      };
+      if (joined.includes('list devices')) {
+        writeJson({ result: { devices: [{ identifier: 'UDID-1', connectionProperties: { tunnelState: 'connected', pairingState: 'paired' }, deviceProperties: { name: 'Test' }, hardwareProperties: { productType: 'iPhone17,1' } }] } });
+        return makeReturn(0);
+      }
+      if (joined.includes('info processes')) { writeJson({ result: { runningProcesses: [] } }); return makeReturn(0); } // not running -> launch
+      if (joined.includes('process launch')) { return makeReturn(0); }
+      if (joined.includes('info details')) { writeJson({ result: { connectionProperties: { tunnelIPAddress: 'fd00::1' } } }); return makeReturn(0); }
+      if (joined.includes('copy from')) {
+        const fi = args.indexOf('--destination');
+        if (fi !== -1 && args[fi + 1]) writeFileSync(args[fi + 1]!, 'BOOT-TOK\n');
+        return makeReturn(0);
+      }
+      return makeReturn(1, '', 'unexpected ' + joined);
+    }) as SpawnImpl;
+
+    const r = await bootstrapTunnel({
+      udid: 'UDID-1',
+      bundleId: 'com.test.app',
+      launchEnv: { BH_ENABLE_IOS_QA_BRIDGE: '1' },
+      spawnImpl: spawn,
+      resolveImpl: async () => ['fd00::1'],
+      fetchImpl: (async (url: unknown) => {
+        const u = String(url);
+        if (u.endsWith('/healthz')) return new Response('{"version":"1.0.0"}', { status: 200 });
+        if (u.endsWith('/auth/rotate')) return new Response('{"ok":true}', { status: 200 });
+        return new Response('nope', { status: 404 });
+      }) as typeof fetch,
+      startupTimeoutMs: 1_000,
+    });
+
+    expect(r.ok).toBe(true);
+    const launchCall = calls.find((a) => a.join(' ').includes('process launch'));
+    expect(launchCall).toBeDefined();
+    const i = launchCall!.indexOf('--environment-variables');
+    expect(i).toBeGreaterThan(-1);
+    expect(JSON.parse(launchCall![i + 1]!)).toEqual({ BH_ENABLE_IOS_QA_BRIDGE: '1' });
+  });
+});