@buoy-gg/mcp 7.0.20 → 7.0.22

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.
package/dist/cli.js CHANGED
@@ -37,7 +37,12 @@ var import_sync_broker = require("@buoy-gg/sync-broker");
37
37
  var BROKER_PORT = import_sync_broker.DEFAULT_BROKER_PORT;
38
38
  var BROKER_URL = `http://localhost:${BROKER_PORT}`;
39
39
  var DEFAULT_ACTION_TIMEOUT_MS = 15e3;
40
+ var UNRESPONSIVE_ACTION_TIMEOUT_MS = 2500;
41
+ var UNRESPONSIVE_SNAPSHOT_TIMEOUT_MS = 1200;
42
+ var RECOVERY_PROBE_MS = 2e4;
43
+ var UNRESPONSIVE_AFTER_MS = 3e4;
40
44
  var DEFAULT_SNAPSHOT_TIMEOUT_MS = 5e3;
45
+ var DEVICE_RETENTION_MS = 15 * 6e4;
41
46
  var DASHBOARD_NAME = "Dashboard";
42
47
  var ALL_DEVICES = "All";
43
48
 
@@ -62,6 +67,16 @@ var SyncClient = class {
62
67
  this.pendingActions = /* @__PURE__ */ new Map();
63
68
  /** Callbacks waiting for the next snapshot of `${deviceId}::${toolId}`. */
64
69
  this.snapshotWaiters = /* @__PURE__ */ new Map();
70
+ /** When each cached snapshot arrived, keyed by snapKey (staleness reporting). */
71
+ this.snapshotAt = /* @__PURE__ */ new Map();
72
+ /** Liveness per device — see {@link noteTimeout}. */
73
+ this.health = /* @__PURE__ */ new Map();
74
+ /** Dev-server origin per device — see {@link noteDevServerOrigin}. */
75
+ this.devServerOrigins = /* @__PURE__ */ new Map();
76
+ /** Devices whose origin probe already ran (a production bundle has none). */
77
+ this.originProbed = /* @__PURE__ */ new Set();
78
+ /** Last time each device appeared in the broker's device list. */
79
+ this.lastSeen = /* @__PURE__ */ new Map();
65
80
  this.port = opts.port ?? BROKER_PORT;
66
81
  this.url = opts.url ?? (opts.port ? `http://localhost:${opts.port}` : BROKER_URL);
67
82
  this.actionTimeoutMs = opts.actionTimeoutMs ?? DEFAULT_ACTION_TIMEOUT_MS;
@@ -121,13 +136,19 @@ var SyncClient = class {
121
136
  });
122
137
  socket.on("devtool-capabilities", (msg) => {
123
138
  this.capabilities.set(msg.persistentDeviceId, msg);
139
+ void this.learnDevServerOrigin(msg.persistentDeviceId);
124
140
  });
125
141
  socket.on("all-devices-update", (list) => {
126
- if (Array.isArray(list)) this.devices = list;
142
+ if (!Array.isArray(list)) return;
143
+ this.devices = list;
144
+ const now = Date.now();
145
+ for (const d of list) if (d.deviceId) this.lastSeen.set(d.deviceId, now);
146
+ this.pruneStaleDevices();
127
147
  });
128
148
  socket.on("devtool-sync", (msg) => {
129
149
  const key = snapKey(msg.persistentDeviceId, msg.toolId);
130
150
  this.snapshots.set(key, msg.payload);
151
+ this.snapshotAt.set(key, Date.now());
131
152
  const waiters = this.snapshotWaiters.get(key);
132
153
  if (waiters && waiters.length) {
133
154
  this.snapshotWaiters.set(key, []);
@@ -143,6 +164,37 @@ var SyncClient = class {
143
164
  });
144
165
  });
145
166
  }
167
+ /**
168
+ * Drop everything remembered about devices that have been gone a while.
169
+ *
170
+ * Deliberately NOT on disconnect: a crashed app often drops off the broker,
171
+ * and its cached console snapshot is the crash evidence — throwing that away
172
+ * the instant it disconnects would delete the answer. But device ids change
173
+ * with every simulator erase and fresh dev client, and each one can hold
174
+ * multi-MB snapshots, so a long-lived server has to forget eventually.
175
+ */
176
+ pruneStaleDevices() {
177
+ const cutoff = Date.now() - DEVICE_RETENTION_MS;
178
+ for (const [deviceId, seen] of this.lastSeen) {
179
+ if (seen > cutoff) continue;
180
+ if (this.devices.some((d) => d.deviceId === deviceId)) continue;
181
+ this.lastSeen.delete(deviceId);
182
+ this.capabilities.delete(deviceId);
183
+ this.health.delete(deviceId);
184
+ this.devServerOrigins.delete(deviceId);
185
+ this.originProbed.delete(deviceId);
186
+ const prefix = `${deviceId}::`;
187
+ for (const key of [...this.snapshots.keys()]) {
188
+ if (key.startsWith(prefix)) {
189
+ this.snapshots.delete(key);
190
+ this.snapshotAt.delete(key);
191
+ }
192
+ }
193
+ for (const key of [...this.watching.keys()]) {
194
+ if (key.startsWith(prefix)) this.watching.delete(key);
195
+ }
196
+ }
197
+ }
146
198
  /** Disconnect and tear down a spawned broker, if any. */
147
199
  async stop() {
148
200
  for (const [, p] of this.pendingActions) {
@@ -150,6 +202,8 @@ var SyncClient = class {
150
202
  p.reject(new Error("SyncClient stopped"));
151
203
  }
152
204
  this.pendingActions.clear();
205
+ for (const [, waiters] of this.snapshotWaiters) for (const w of waiters) w(void 0);
206
+ this.snapshotWaiters.clear();
153
207
  this.socket?.close();
154
208
  this.socket = null;
155
209
  if (this.broker) {
@@ -193,11 +247,19 @@ var SyncClient = class {
193
247
  resolveDevice(deviceId) {
194
248
  if (deviceId) return deviceId;
195
249
  const connected = this.connectedDevices();
196
- if (connected.length === 1) return connected[0].deviceId;
197
250
  if (connected.length === 0) {
198
251
  throw new Error("No devices are connected to the broker. Open the app with Buoy devtools running.");
199
252
  }
200
- const list = connected.map((d) => `${d.deviceName} (${d.deviceId})`).join(", ");
253
+ if (connected.length === 1) return connected[0].deviceId;
254
+ const usable = connected.filter((d) => d.tools.length > 0);
255
+ if (usable.length === 1) return usable[0].deviceId;
256
+ if (usable.length === 0) {
257
+ const names = connected.map((d) => d.deviceName).join(", ");
258
+ throw new Error(
259
+ `Connected devices (${names}) are not exposing any Buoy tools yet. Make sure the app is in the foreground with devtools mounted, then retry.`
260
+ );
261
+ }
262
+ const list = usable.map((d) => `${d.deviceName} (${d.deviceId})`).join(", ");
201
263
  throw new Error(`Multiple devices connected \u2014 specify deviceId. Connected: ${list}`);
202
264
  }
203
265
  /**
@@ -222,6 +284,136 @@ var SyncClient = class {
222
284
  return target;
223
285
  }
224
286
  // ----------------------------------------------------------------
287
+ // Device liveness
288
+ // ----------------------------------------------------------------
289
+ healthOf(deviceId) {
290
+ let h = this.health.get(deviceId);
291
+ if (!h) {
292
+ h = { consecutiveTimeouts: 0, lastOkAt: 0, lastFullBudgetAt: 0 };
293
+ this.health.set(deviceId, h);
294
+ }
295
+ return h;
296
+ }
297
+ /** The device answered — it's alive. */
298
+ noteOk(deviceId) {
299
+ const h = this.healthOf(deviceId);
300
+ h.consecutiveTimeouts = 0;
301
+ h.lastOkAt = Date.now();
302
+ }
303
+ /**
304
+ * The device didn't answer. A crashed app never answers again, so after the
305
+ * first miss we stop paying the full timeout on every subsequent call.
306
+ */
307
+ noteTimeout(deviceId) {
308
+ this.healthOf(deviceId).consecutiveTimeouts += 1;
309
+ }
310
+ /** True once a device has missed a call and hasn't answered since. */
311
+ isUnresponsive(deviceId) {
312
+ return this.healthOf(deviceId).consecutiveTimeouts > 0;
313
+ }
314
+ /**
315
+ * Full timeout normally; a short one for a device already not answering.
316
+ *
317
+ * With one escape hatch: every RECOVERY_PROBE_MS a flagged device gets the
318
+ * full budget again. Without it the flag is self-sustaining — a device whose
319
+ * work legitimately takes longer than the short budget (a big snapshot, a
320
+ * slow simulator) can never answer, so it can never clear the flag, and every
321
+ * later reply is a confident "most likely CRASHED" about a healthy app.
322
+ */
323
+ timeoutFor(deviceId) {
324
+ if (!this.isUnresponsive(deviceId)) return this.actionTimeoutMs;
325
+ const h = this.healthOf(deviceId);
326
+ if (Date.now() - h.lastFullBudgetAt > RECOVERY_PROBE_MS) {
327
+ h.lastFullBudgetAt = Date.now();
328
+ return this.actionTimeoutMs;
329
+ }
330
+ return UNRESPONSIVE_ACTION_TIMEOUT_MS;
331
+ }
332
+ /**
333
+ * Plain-language diagnosis for a device that isn't answering, including the
334
+ * app's last words if the console tool got a crash flush out in time.
335
+ *
336
+ * A silent timeout is the worst possible answer here: the app is on a redbox,
337
+ * the reason is already captured, and without this the caller reads "timed
338
+ * out" and keeps retrying the same dead device.
339
+ */
340
+ unresponsiveNote(deviceId) {
341
+ const h = this.healthOf(deviceId);
342
+ const since = h.lastOkAt ? Math.round((Date.now() - h.lastOkAt) / 1e3) : null;
343
+ const crashed = h.consecutiveTimeouts >= 2 || h.lastOkAt === 0 || Date.now() - h.lastOkAt > UNRESPONSIVE_AFTER_MS;
344
+ const lines = [
345
+ `The app on "${deviceId}" is not responding (${h.consecutiveTimeouts} call${h.consecutiveTimeouts === 1 ? "" : "s"} unanswered${since != null ? `, last answered ${since}s ago` : ""}).`,
346
+ crashed ? "It is most likely CRASHED \u2014 a fatal render error unmounts the React tree, and the dev tools live in that tree, so nothing on-device can answer until the app is reloaded." : "It may be busy, mid-reload, or crashing."
347
+ ];
348
+ const fatal = this.lastFatalFor(deviceId);
349
+ if (fatal) lines.push(`Last captured error: ${fatal}`);
350
+ lines.push(
351
+ fatal ? "Read the crash state with get_console / get_storage (served from the last sync), then reload_app to recover." : "Try get_console (served from the last sync) for the app's last words, then reload_app to recover."
352
+ );
353
+ return lines.join(" ");
354
+ }
355
+ /**
356
+ * The newest error line the console tool synced for this device, if any.
357
+ * Reads the cached console snapshot rather than the device (which, by the
358
+ * time this matters, is not answering).
359
+ */
360
+ lastFatalFor(deviceId) {
361
+ const snap = this.snapshots.get(snapKey(deviceId, "console"));
362
+ const entries = snap?.entries;
363
+ if (!Array.isArray(entries)) return null;
364
+ for (let i = entries.length - 1; i >= 0; i--) {
365
+ const e = entries[i];
366
+ if (e?.fatal || e?.level === "error") {
367
+ const msg = (e.message ?? "").split("\n")[0];
368
+ if (msg) return msg.length > 300 ? `${msg.slice(0, 300)}\u2026` : msg;
369
+ }
370
+ }
371
+ return null;
372
+ }
373
+ /** Age of the cached snapshot for a tool, in ms (null = never received). */
374
+ snapshotAgeMs(toolId, deviceId) {
375
+ const at = this.snapshotAt.get(snapKey(deviceId, toolId));
376
+ return at ? Date.now() - at : null;
377
+ }
378
+ /**
379
+ * Remember where a device's bundle came from, reported by core's `app`
380
+ * adapter while the app is healthy. It is the only way to reload an app that
381
+ * has crashed: the in-app reload needs the React tree that the crash
382
+ * destroyed, but the dev server's reload is handled natively.
383
+ */
384
+ noteDevServerOrigin(deviceId, origin) {
385
+ if (typeof origin === "string" && /^https?:\/\//.test(origin)) {
386
+ this.devServerOrigins.set(deviceId, origin);
387
+ }
388
+ }
389
+ /**
390
+ * Ask a freshly-announced device where its bundle came from, once.
391
+ *
392
+ * Deliberately eager: this is the piece of information that can ONLY be
393
+ * collected while the app is healthy, and is only ever needed after it has
394
+ * stopped answering. One cheap round-trip per device buys the ability to
395
+ * recover a crashed app instead of telling the caller to go do it by hand.
396
+ */
397
+ async learnDevServerOrigin(deviceId) {
398
+ if (this.devServerOrigins.has(deviceId) || this.originProbed.has(deviceId)) return;
399
+ this.originProbed.add(deviceId);
400
+ const hasApp = this.capabilities.get(deviceId)?.tools?.some((t) => t.toolId === "app");
401
+ if (!hasApp) return;
402
+ try {
403
+ const { payload } = await this.getSnapshot("app", deviceId, { timeoutMs: 4e3, silent: true });
404
+ this.noteDevServerOrigin(deviceId, payload?.devServerOrigin);
405
+ } catch {
406
+ }
407
+ }
408
+ /** The dev-server origin last reported by this device, if any. */
409
+ devServerOrigin(deviceId) {
410
+ const known = this.devServerOrigins.get(deviceId);
411
+ if (known) return known;
412
+ const snap = this.snapshots.get(snapKey(deviceId, "app"));
413
+ if (typeof snap?.devServerOrigin === "string") return snap.devServerOrigin;
414
+ return null;
415
+ }
416
+ // ----------------------------------------------------------------
225
417
  // Watch + snapshots
226
418
  // ----------------------------------------------------------------
227
419
  /**
@@ -292,13 +484,36 @@ var SyncClient = class {
292
484
  const target = this.resolveDevice(deviceId);
293
485
  const key = snapKey(target, toolId);
294
486
  this.ensureWatching(toolId, target);
295
- const next = this.waitForSnapshot(key, opts.timeoutMs ?? DEFAULT_SNAPSHOT_TIMEOUT_MS);
487
+ let budget = opts.timeoutMs ?? DEFAULT_SNAPSHOT_TIMEOUT_MS;
488
+ if (opts.timeoutMs === void 0 && this.isUnresponsive(target)) {
489
+ const h = this.healthOf(target);
490
+ if (Date.now() - h.lastFullBudgetAt > RECOVERY_PROBE_MS) {
491
+ h.lastFullBudgetAt = Date.now();
492
+ } else {
493
+ budget = UNRESPONSIVE_SNAPSHOT_TIMEOUT_MS;
494
+ }
495
+ }
496
+ const next = this.waitForSnapshot(key, budget);
296
497
  this.socket?.emit("request-devtool-state", { toolId, targetDeviceId: target });
297
498
  const payload = await next;
298
- if (payload !== void 0) return { deviceId: target, payload };
499
+ if (payload !== void 0) {
500
+ if (!opts.silent) this.noteOk(target);
501
+ return { deviceId: target, payload };
502
+ }
503
+ if (!opts.silent) this.noteTimeout(target);
299
504
  const cached = this.snapshots.get(key);
300
- if (cached !== void 0) return { deviceId: target, payload: cached };
301
- throw new Error(`Timed out waiting for "${toolId}" state from ${target} (is the tool active?)`);
505
+ if (cached !== void 0) {
506
+ return { deviceId: target, payload: cached, stale: this.staleNote(toolId, target) };
507
+ }
508
+ throw new Error(
509
+ `Timed out waiting for "${toolId}" state from ${target} (is the tool active?). ${this.unresponsiveNote(target)}`
510
+ );
511
+ }
512
+ /** Header line for data served from cache because the device went quiet. */
513
+ staleNote(toolId, deviceId) {
514
+ const age = this.snapshotAgeMs(toolId, deviceId);
515
+ const ageText = age == null ? "an earlier sync" : `${Math.round(age / 1e3)}s ago`;
516
+ return `\u26A0\uFE0F Served from the last sync (${ageText}) \u2014 ${this.unresponsiveNote(deviceId)}`;
302
517
  }
303
518
  waitForSnapshot(key, timeoutMs) {
304
519
  return new Promise((resolve) => {
@@ -327,11 +542,16 @@ var SyncClient = class {
327
542
  const target = deviceId === ALL_DEVICES ? ALL_DEVICES : this.resolveDevice(deviceId);
328
543
  if (!this.socket) throw new Error("SyncClient is not connected");
329
544
  const requestId = (0, import_node_crypto.randomUUID)();
330
- const timeoutMs = opts.timeoutMs ?? this.actionTimeoutMs;
545
+ const timeoutMs = opts.timeoutMs ?? this.timeoutFor(target);
331
546
  const result = await new Promise((resolve, reject) => {
332
547
  const timer = setTimeout(() => {
333
548
  this.pendingActions.delete(requestId);
334
- reject(new Error(`Action "${toolId}.${action}" timed out after ${timeoutMs}ms`));
549
+ if (target !== ALL_DEVICES && !opts.silent) this.noteTimeout(target);
550
+ reject(
551
+ new Error(
552
+ `Action "${toolId}.${action}" got no answer in ${timeoutMs}ms. ` + (target === ALL_DEVICES ? "No device responded." : this.unresponsiveNote(target))
553
+ )
554
+ );
335
555
  }, timeoutMs);
336
556
  this.pendingActions.set(requestId, { resolve, reject, timer });
337
557
  this.socket.emit("devtool-action", {
@@ -342,6 +562,7 @@ var SyncClient = class {
342
562
  params: params ?? {}
343
563
  });
344
564
  });
565
+ if (target !== ALL_DEVICES && !opts.silent) this.noteOk(target);
345
566
  if (!result.ok) {
346
567
  throw new Error(result.error || `Action "${toolId}.${action}" failed on device`);
347
568
  }
@@ -373,7 +594,7 @@ var delay = (ms) => new Promise((r) => setTimeout(r, ms));
373
594
 
374
595
  // src/server.ts
375
596
  var import_mcp = require("@modelcontextprotocol/sdk/server/mcp.js");
376
- var import_zod19 = require("zod");
597
+ var import_zod22 = require("zod");
377
598
 
378
599
  // src/tools/generic.ts
379
600
  var import_zod = require("zod");
@@ -384,7 +605,7 @@ var import_node_os = __toESM(require("os"));
384
605
  var import_node_path = __toESM(require("path"));
385
606
 
386
607
  // src/version.ts
387
- var VERSION = "7.0.20" ? "7.0.20" : "0.0.0-dev";
608
+ var VERSION = "7.0.22" ? "7.0.22" : "0.0.0-dev";
388
609
  var PACKAGE_NAME = "@buoy-gg/mcp";
389
610
 
390
611
  // src/update/checkForUpdate.ts
@@ -691,6 +912,7 @@ function rankBatchFromIndex(index, batchId, casesOrder, failures = []) {
691
912
  deltaUiFps: null,
692
913
  deltaJsFps: null,
693
914
  renderCommits: num(median?.renderCommits),
915
+ renderWasted: median?.renderWasted,
694
916
  topRenderers: median?.topRenderers ?? []
695
917
  };
696
918
  return { ...base, flags: failed ? [] : computeFlags(base) };
@@ -781,13 +1003,18 @@ function formatRankingMarkdown(ranking) {
781
1003
  );
782
1004
  }
783
1005
  const renderLines = ranking.rows.filter((r) => !r.failed && (r.renderCommits != null || r.topRenderers.length > 0)).map((r) => {
784
- const top = r.topRenderers.map((t) => `${t.name} \xD7${t.renders} (${t.totalMs}ms)`).join(", ");
1006
+ const top = r.topRenderers.map(
1007
+ (t) => `${t.name} \xD7${t.renders} (${t.totalMs}ms${t.cause ? `, ${t.cause}` : ""})`
1008
+ ).join(", ");
785
1009
  const commits = r.renderCommits != null ? `${r.renderCommits} commits` : "";
786
- return `- **${r.caseName}** \u2014 ${[commits, top].filter(Boolean).join(": ")}`;
1010
+ const wasted = r.renderWasted != null ? `${r.renderWasted} wasted` : "";
1011
+ return `- **${r.caseName}** \u2014 ${[commits, wasted, top].filter(Boolean).join(": ")}`;
787
1012
  });
788
1013
  if (renderLines.length) {
789
1014
  lines.push("");
790
- lines.push("**Renders (median run)** \u2014 heaviest components by render self-time:");
1015
+ lines.push(
1016
+ "**Renders (median run)** \u2014 `wasted` = parent cascades + identity-only prop churn (removable). Cause per component: `propsUnstable` means a parent handed it a fresh function/object with unchanged values \u2014 fix upstream."
1017
+ );
791
1018
  lines.push(...renderLines);
792
1019
  }
793
1020
  lines.push("");
@@ -810,7 +1037,20 @@ function compareReportStats(a, b) {
810
1037
  cpuUsage: ch(a.stats.cpuUsage, b.stats.cpuUsage),
811
1038
  memoryUsage: ch(a.stats.memoryUsage, b.stats.memoryUsage),
812
1039
  jsJankFrames: { a: a.stats.jsJankFrames, b: b.stats.jsJankFrames, delta: b.stats.jsJankFrames - a.stats.jsJankFrames },
813
- uiJankFrames: { a: a.stats.uiJankFrames, b: b.stats.uiJankFrames, delta: b.stats.uiJankFrames - a.stats.uiJankFrames }
1040
+ uiJankFrames: { a: a.stats.uiJankFrames, b: b.stats.uiJankFrames, delta: b.stats.uiJankFrames - a.stats.uiJankFrames },
1041
+ ...compareRenders(a.renders, b.renders)
1042
+ };
1043
+ }
1044
+ function compareRenders(a, b) {
1045
+ if (!a && !b) return null;
1046
+ const n = (v) => v ?? 0;
1047
+ return {
1048
+ renders: {
1049
+ total: { a: n(a?.totalRenders), b: n(b?.totalRenders), delta: n(b?.totalRenders) - n(a?.totalRenders) },
1050
+ commits: { a: n(a?.totalCommits), b: n(b?.totalCommits), delta: n(b?.totalCommits) - n(a?.totalCommits) },
1051
+ wasted: { a: n(a?.wastedRenders), b: n(b?.wastedRenders), delta: n(b?.wastedRenders) - n(a?.wastedRenders) },
1052
+ causes: { a: a?.causes ?? {}, b: b?.causes ?? {} }
1053
+ }
814
1054
  };
815
1055
  }
816
1056
  function round(n, d) {
@@ -1060,7 +1300,7 @@ ${JSON.stringify(summary, null, 2)}
1060
1300
  async ({ deviceId, maxWaitMs }) => {
1061
1301
  const log = (m) => process.stderr.write(`[buoy-mcp] ${m}
1062
1302
  `);
1063
- const cap = maxWaitMs ?? 6e5;
1303
+ const cap3 = maxWaitMs ?? 6e5;
1064
1304
  try {
1065
1305
  const target = sync.requireProDevice(deviceId);
1066
1306
  const start = Date.now();
@@ -1068,7 +1308,7 @@ ${JSON.stringify(summary, null, 2)}
1068
1308
  let idleStreak = 0;
1069
1309
  let lastCount = -1;
1070
1310
  let resolved = null;
1071
- while (Date.now() - start < cap) {
1311
+ while (Date.now() - start < cap3) {
1072
1312
  let snap2;
1073
1313
  try {
1074
1314
  snap2 = await readPerfSnapshot(sync, target);
@@ -1153,7 +1393,7 @@ ${JSON.stringify(ranking, null, 2)}
1153
1393
  server.registerTool(
1154
1394
  "compare_reports",
1155
1395
  {
1156
- description: "Compare two saved benchmark runs head-to-head (avg + p95 for FPS/CPU/memory, jank, with deltas).",
1396
+ description: "Compare two saved benchmark runs head-to-head: avg + p95 for FPS/CPU/memory, jank, AND render counts \u2014 total, commits, and wasted (parent cascades + identity-only prop churn) \u2014 each with deltas. Use this to prove a memoization or re-render fix actually landed: the wasted-render delta is the number that says so.",
1157
1397
  inputSchema: {
1158
1398
  reportIdA: import_zod2.z.string().describe("First (baseline) report id."),
1159
1399
  reportIdB: import_zod2.z.string().describe("Second report id to compare against the baseline."),
@@ -1171,8 +1411,8 @@ ${JSON.stringify(ranking, null, 2)}
1171
1411
  if (!b) return fail2(`Report "${reportIdB}" not found.`);
1172
1412
  return json2(
1173
1413
  compareReportStats(
1174
- { name: a.metadata.name, stats: a.stats },
1175
- { name: b.metadata.name, stats: b.stats }
1414
+ { name: a.metadata.name, stats: a.stats, renders: a.renders },
1415
+ { name: b.metadata.name, stats: b.stats, renders: b.renders }
1176
1416
  )
1177
1417
  );
1178
1418
  } catch (e) {
@@ -1290,7 +1530,7 @@ function registerEventsTools(server, sync) {
1290
1530
  server.registerTool(
1291
1531
  "get_events",
1292
1532
  {
1293
- description: "The UNIVERSAL, token-bounded activity reader for what's happening in the app across ALL Buoy tools. The events tool aggregates every source \u2014 network requests, redux/zustand/jotai actions, react-query query/mutation updates, storage writes, route changes, renders \u2014 so filter to one with `sources`: ['network'] lists HTTP requests, ['redux'] dispatched actions, ['react-query'] query updates, ['storage'] writes, ['route'] navigation. Returns a compact one-line summary per event (source \xB7 status \xB7 title \xB7 timing) by DEFAULT \u2014 not the heavy raw payloads \u2014 so it's cheap. Re-call with includeData:true, a bigger limit, or specific sources only when you need detail. PREFER this over get_snapshot (and over reading source code) for any 'what happened' / 'show me the requests/actions' question; get_snapshot is the raw heavy current-state escape hatch.",
1533
+ description: "The UNIVERSAL, token-bounded activity reader for what's happening in the app across ALL Buoy tools. Check this timeline BEFORE driving the UI to reproduce a bug \u2014 the failure is usually already recorded here. The events tool aggregates every source \u2014 network requests, redux/zustand/jotai actions, react-query query/mutation updates, storage writes, route changes, renders \u2014 so filter to one with `sources`: ['network'] lists HTTP requests, ['redux'] dispatched actions, ['react-query'] query updates, ['storage'] writes, ['route'] navigation. Returns a compact one-line summary per event (source \xB7 status \xB7 title \xB7 timing) by DEFAULT \u2014 not the heavy raw payloads \u2014 so it's cheap. Re-call with includeData:true, a bigger limit, or specific sources only when you need detail. PREFER this over get_snapshot (and over reading source code) for any 'what happened' / 'show me the requests/actions' question; get_snapshot is the raw heavy current-state escape hatch.",
1294
1534
  inputSchema: {
1295
1535
  limit: import_zod3.z.number().optional().describe("Most-recent N events to include. Default 25."),
1296
1536
  sources: import_zod3.z.array(import_zod3.z.string()).optional().describe('Only these sources. Coarse names accepted: "network", "redux", "react-query", "storage", "route", "render", "zustand", "jotai". Omit for all.'),
@@ -1455,7 +1695,7 @@ function registerNetworkTools(server, sync) {
1455
1695
  server.registerTool(
1456
1696
  "get_network_requests",
1457
1697
  {
1458
- description: "List captured HTTP requests WITH their request ids, plus which ones are pinned/saved. Use this (not get_events) when you need to ACT on a request \u2014 every network action is keyed by the id, and get_events does not emit ids. Compact by default: one line per request (id \xB7 status \xB7 method \xB7 path \xB7 timing \xB7 size). Set flagged:'pinned'|'saved'|'any' to read the developer's KEPT requests \u2014 those are snapshots that outlive Clear, the 500-request cap and app restarts, so treat a pinned request as 'a human flagged this as the problem'. Filter with status:'errors' to go straight to failures, or `pattern` to match a URL substring. includeBodies:true adds truncated request/response payloads (heavier).",
1698
+ description: "List captured HTTP requests WITH their request ids, plus which ones are pinned/saved. API data wrong or calls failing \u2192 read the actual request+response bodies here instead of guessing from code; capture starts at app launch, so boot-time requests (session bootstrap, first queries) are included. Use this (not get_events) when you need to ACT on a request \u2014 every network action is keyed by the id, and get_events does not emit ids. Compact by default: one line per request (id \xB7 status \xB7 method \xB7 path \xB7 timing \xB7 size). Set flagged:'pinned'|'saved'|'any' to read the developer's KEPT requests \u2014 those are snapshots that outlive Clear, the 500-request cap and app restarts, so treat a pinned request as 'a human flagged this as the problem'. Filter with status:'errors' to go straight to failures, or `pattern` to match a URL substring. includeBodies:true adds truncated request/response payloads (heavier).",
1459
1699
  inputSchema: {
1460
1700
  limit: import_zod4.z.number().optional().describe("Most-recent N requests. Default 25."),
1461
1701
  pattern: import_zod4.z.string().optional().describe("Only requests whose URL contains this substring (case-insensitive)."),
@@ -2172,7 +2412,7 @@ function registerStorageTools(server, sync) {
2172
2412
  server.registerTool(
2173
2413
  "get_storage",
2174
2414
  {
2175
- description: "Read the app's storage keys across ALL backends: AsyncStorage, MMKV (per instance), and Expo SecureStore. Compact by DEFAULT \u2014 keys only \u2014 so it's token-cheap; pass includeValues:true for the values (truncated), or a `pattern` substring to narrow. Buoy's own internal keys are already filtered out. Use for 'what's in storage / read this key'.",
2415
+ description: "Read the app's storage keys across ALL backends: AsyncStorage, MMKV (per instance), and Expo SecureStore. Data wrong or missing after an app restart \u2192 read the persisted values here FIRST \u2014 the bug is usually a stale, missing, or malformed stored value, and reading it beats hunting for it in code. Compact by DEFAULT \u2014 keys only \u2014 so it's token-cheap; pass includeValues:true for the values (truncated), or a `pattern` substring to narrow. Buoy's own internal keys are already filtered out.",
2176
2416
  inputSchema: {
2177
2417
  pattern: import_zod9.z.string().optional().describe("Only keys containing this substring (case-insensitive)."),
2178
2418
  includeValues: import_zod9.z.boolean().optional().describe("Include each key's value (truncated). Default false (keys only)."),
@@ -2185,13 +2425,13 @@ function registerStorageTools(server, sync) {
2185
2425
  const target = sync.requireProDevice(args.deviceId);
2186
2426
  const pattern = args.pattern?.toLowerCase();
2187
2427
  const limit = typeof args.limit === "number" && args.limit > 0 ? args.limit : 100;
2188
- const matches = (k) => !pattern || k.toLowerCase().includes(pattern);
2428
+ const matches2 = (k) => !pattern || k.toLowerCase().includes(pattern);
2189
2429
  const sections = [];
2190
2430
  const allKeys = await sync.callAction(STORAGE_TOOL_ID, "async.getAllKeys", {}, target);
2191
2431
  if (!Array.isArray(allKeys)) {
2192
2432
  return fail9("No storage result from the device. The storage tool may be inactive.");
2193
2433
  }
2194
- const asyncKeys = allKeys.filter(matches).slice(0, limit);
2434
+ const asyncKeys = allKeys.filter(matches2).slice(0, limit);
2195
2435
  if (asyncKeys.length > 0) {
2196
2436
  let lines;
2197
2437
  if (args.includeValues) {
@@ -2206,7 +2446,7 @@ ${lines.join("\n")}`);
2206
2446
  }
2207
2447
  const mmkv = await sync.callAction(STORAGE_TOOL_ID, "mmkv.snapshot", {}, target).catch(() => null);
2208
2448
  for (const inst of mmkv ?? []) {
2209
- const entries = inst.entries.filter((e) => matches(e.key)).slice(0, limit);
2449
+ const entries = inst.entries.filter((e) => matches2(e.key)).slice(0, limit);
2210
2450
  if (entries.length === 0) continue;
2211
2451
  const lines = entries.map(
2212
2452
  (e) => args.includeValues ? `- \`${e.key}\` (${e.valueType}) = ${truncate2(e.value)}` : `- \`${e.key}\``
@@ -2216,7 +2456,7 @@ ${lines.join("\n")}`);
2216
2456
  ${lines.join("\n")}`);
2217
2457
  }
2218
2458
  const secure = await sync.callAction(STORAGE_TOOL_ID, "secure.keys", {}, target).catch(() => null);
2219
- const secureKeys = (secure ?? []).filter((d) => matches(d.key)).slice(0, limit);
2459
+ const secureKeys = (secure ?? []).filter((d) => matches2(d.key)).slice(0, limit);
2220
2460
  if (secureKeys.length > 0) {
2221
2461
  const lines = await Promise.all(
2222
2462
  secureKeys.map(async (d) => {
@@ -2373,7 +2613,7 @@ function registerConsoleTools(server, sync) {
2373
2613
  server.registerTool(
2374
2614
  "get_console",
2375
2615
  {
2376
- description: "Read the app's captured console logs (console.log/warn/error/\u2026) as a compact tail \u2014 time \xB7 level \xB7 message. Token-cheap (uses each entry's pre-rendered message, not raw args). Filter by minimum level (e.g. level:'warning' for warnings+errors) or a message substring. Use for 'what's logging / any console errors'. For full args/stack of an entry, fall back to get_snapshot('console').",
2616
+ description: "Read the app's captured console logs (console.log/warn/error/\u2026) as a compact tail \u2014 time \xB7 level \xB7 message. App crashing or erroring \u2192 read this FIRST: it captures crashes and errors including FATAL uncaught JS errors (tagged [FATAL], hooked from RN's global error handler), so a dead app's last words are here, not in the code. Token-cheap (uses each entry's pre-rendered message, not raw args). Filter by minimum level (e.g. level:'warning' for warnings+errors) or a message substring. For full args/stack of an entry, fall back to get_snapshot('console').",
2377
2617
  inputSchema: {
2378
2618
  level: import_zod11.z.enum(["verbose", "info", "warning", "error"]).optional().describe("Minimum level to include (severity threshold). e.g. 'warning' = warnings + errors. Default all."),
2379
2619
  limit: import_zod11.z.number().optional().describe("Most-recent N entries. Default 50."),
@@ -2384,7 +2624,10 @@ function registerConsoleTools(server, sync) {
2384
2624
  async (args) => {
2385
2625
  try {
2386
2626
  const target = sync.requireProDevice(args.deviceId);
2387
- const { payload } = await sync.getSnapshot(CONSOLE_TOOL_ID, target);
2627
+ const { payload, stale } = await sync.getSnapshot(CONSOLE_TOOL_ID, target);
2628
+ const staleHeader = stale ? `${stale}
2629
+
2630
+ ` : "";
2388
2631
  const p = Array.isArray(payload) ? { entries: payload } : payload ?? {};
2389
2632
  const all = Array.isArray(p.entries) ? p.entries : [];
2390
2633
  const minSev = args.level ? SEVERITY[args.level] : 0;
@@ -2396,12 +2639,12 @@ function registerConsoleTools(server, sync) {
2396
2639
  const limit = typeof args.limit === "number" && args.limit > 0 ? args.limit : 50;
2397
2640
  const shown = sorted.slice(-limit);
2398
2641
  if (shown.length === 0) {
2399
- return text10(`No console entries${args.level ? ` at level \u2265 ${args.level}` : ""}${pattern ? ` matching "${args.pattern}"` : ""} (${all.length} total captured).`);
2642
+ return text10(`${staleHeader}No console entries${args.level ? ` at level \u2265 ${args.level}` : ""}${pattern ? ` matching "${args.pattern}"` : ""} (${all.length} total captured).`);
2400
2643
  }
2401
2644
  const lines = shown.map((e) => `${hhmmss(e.timestamp)} [${TAG[e.level] ?? e.level}] ${e.message}`);
2402
2645
  const levelNote = args.level ? `, level \u2265 ${args.level}` : "";
2403
2646
  const header = `**Console \u2014 ${all.length} entries (showing ${shown.length}${levelNote}):**`;
2404
- return text10(`${header}
2647
+ return text10(`${staleHeader}${header}
2405
2648
 
2406
2649
  \`\`\`
2407
2650
  ${lines.join("\n")}
@@ -2425,13 +2668,19 @@ function kb(bytes) {
2425
2668
  return bytes < 1024 ? `${bytes} B` : `${(bytes / 1024).toFixed(1)} KB`;
2426
2669
  }
2427
2670
  function formatOutcome(outcome) {
2428
- const lines = Object.entries(outcome.results).map(([sid, r]) => {
2429
- const skips = r.skipped.slice(0, 3).map((s) => `
2671
+ const lines = Object.entries(outcome.results).map(([sid, r2]) => {
2672
+ const skips = r2.skipped.slice(0, 3).map((s) => `
2430
2673
  \xB7 ${s.id}: ${s.reason}`).join("");
2431
- const warns = r.warnings.slice(0, 3).map((w) => `
2674
+ const warns = r2.warnings.slice(0, 3).map((w) => `
2432
2675
  \u26A0 ${w}`).join("");
2433
- return ` ${r.ok ? "\u2713" : "\u2715"} ${sid} \u2014 applied ${r.applied}${r.skipped.length ? `, skipped ${r.skipped.length}` : ""}${skips}${warns}`;
2676
+ return ` ${r2.ok ? "\u2713" : "\u2715"} ${sid} \u2014 applied ${r2.applied}${r2.skipped.length ? `, skipped ${r2.skipped.length}` : ""}${skips}${warns}`;
2434
2677
  });
2678
+ const r = outcome.route;
2679
+ if (r) {
2680
+ lines.push(
2681
+ r.status === "navigated" ? ` \u2713 route \u2014 navigated to ${r.href}` : r.status === "deferred" ? ` \u21BB route \u2014 ${r.href} will be opened after the reload` : ` \u2715 route \u2014 ${r.error ?? r.status}`
2682
+ );
2683
+ }
2435
2684
  return `Restored "${outcome.snapshotName}" (${outcome.mode}${outcome.willReload ? ", app reloading" : ""}):
2436
2685
  ${lines.join("\n")}`;
2437
2686
  }
@@ -2439,7 +2688,7 @@ function registerTimeMachineTools(server, sync) {
2439
2688
  server.registerTool(
2440
2689
  "get_time_machine",
2441
2690
  {
2442
- description: "List the app's Time Machine state snapshots (restore points) and which state sources (storage/redux/zustand/jotai/react-query) can currently be captured/restored. Use time_machine_action to capture a snapshot before driving a flow, then restore it to re-run the flow from the same state.",
2691
+ description: "List the app's Time Machine state snapshots (restore points) and which state sources (storage/redux/zustand/jotai/react-query) can currently be captured/restored. Each snapshot also records the route it was captured on (shown as `route`), which a restore can navigate back to. Use time_machine_action to capture a snapshot before driving a flow, then restore it to re-run the flow from the same state.",
2443
2692
  inputSchema: {
2444
2693
  deviceId: import_zod12.z.string().optional().describe("Target device id (auto-selected if one is connected).")
2445
2694
  }
@@ -2452,13 +2701,19 @@ function registerTimeMachineTools(server, sync) {
2452
2701
  const snapshots = p.snapshots ?? [];
2453
2702
  const providers = p.providers ?? [];
2454
2703
  const providerLines = providers.map((pr) => {
2455
- const cap = pr.canCapture ? pr.canRestore ? "capture+restore" : "capture only" : "unavailable";
2456
- return `- ${pr.label} (${pr.id}): ${cap}${pr.detail ? ` \u2014 ${pr.detail}` : ""}`;
2704
+ const cap3 = pr.canCapture ? pr.canRestore ? "capture+restore" : "capture only" : "unavailable";
2705
+ return `- ${pr.label} (${pr.id}): ${cap3}${pr.detail ? ` \u2014 ${pr.detail}` : ""}`;
2457
2706
  });
2707
+ if (p.route) {
2708
+ providerLines.push(
2709
+ p.route.available ? `- Route (route): capture+restore \u2014 recorded with every snapshot; pass restoreRoute:true to return to it${p.route.current ? ` (now ${p.route.current})` : ""}` : "- Route (route): unavailable \u2014 no Expo Router on this device, snapshots record no route"
2710
+ );
2711
+ }
2458
2712
  const snapLines = snapshots.map((s) => {
2459
2713
  const warnings = Object.values(s.sources).reduce((n, src) => n + src.warnings.length, 0);
2460
2714
  const desc = s.baseline ? "baseline \u2014 restore wipes to fresh install (always reloads)" : `${kb(s.sizeBytes)} \xB7 ${Object.keys(s.sources).join(", ")}`;
2461
- return `- \`${s.id}\` **${s.name}** \u2014 ${new Date(s.createdAt).toISOString()} \xB7 ${desc}${warnings ? ` \xB7 \u26A0 ${warnings}` : ""}`;
2715
+ const route = s.route ? ` \xB7 route ${s.route.href}` : "";
2716
+ return `- \`${s.id}\` **${s.name}** \u2014 ${new Date(s.createdAt).toISOString()} \xB7 ${desc}${route}${warnings ? ` \xB7 \u26A0 ${warnings}` : ""}`;
2462
2717
  });
2463
2718
  const parts = [
2464
2719
  `**Time Machine \u2014 ${snapshots.length} snapshot(s):**`,
@@ -2477,13 +2732,14 @@ ${formatOutcome(p.lastOutcome)}`);
2477
2732
  server.registerTool(
2478
2733
  "time_machine_action",
2479
2734
  {
2480
- description: "Manage Time Machine state snapshots: capture the app's current state as a named restore point, preview what restoring one would change (per-item added/removed/changed/won't-apply verdicts \u2014 check the blast radius BEFORE restoring), restore one (mode 'live' replaces state in place; 'reload' restores persisted storage then reloads the JS bundle \u2014 baselines always reload), capture_baseline (an empty wipe-to-fresh-install restore point), delete, or rename. The restore result reports per-source applied/skipped counts \u2014 check them rather than assuming success.",
2735
+ description: "Manage Time Machine state snapshots: capture the app's current state as a named restore point, preview what restoring one would change (per-item added/removed/changed/won't-apply verdicts \u2014 check the blast radius BEFORE restoring), restore one (mode 'live' replaces state in place; 'reload' restores persisted storage then reloads the JS bundle \u2014 baselines always reload; pass restoreRoute:true to also return to the screen the snapshot was captured on), capture_baseline (an empty wipe-to-fresh-install restore point), delete, or rename. The restore result reports per-source applied/skipped counts \u2014 check them rather than assuming success.",
2481
2736
  inputSchema: {
2482
2737
  action: import_zod12.z.enum(["capture", "capture_baseline", "preview", "restore", "delete", "rename"]).describe("What to do."),
2483
2738
  id: import_zod12.z.string().optional().describe("Snapshot id (required for restore/delete/rename; from get_time_machine)."),
2484
2739
  name: import_zod12.z.string().optional().describe("Snapshot name (for capture and rename)."),
2485
2740
  mode: import_zod12.z.enum(["live", "reload"]).optional().describe("Restore mode. Default 'live'."),
2486
2741
  excludeKeys: import_zod12.z.array(import_zod12.z.string()).optional().describe('Restore only: item keys to LEAVE UNTOUCHED, as "sourceId::itemKey" (itemKey = the key shown by the preview action, e.g. "query::[\\"pokemon\\"]" or "storage::async:@cart").'),
2742
+ restoreRoute: import_zod12.z.boolean().optional().describe("Restore only: also navigate back to the route the snapshot was captured on (shown as `route` by get_time_machine). With mode 'reload' the navigation happens automatically on the next boot. Default false \u2014 state is restored but the app stays on the current screen."),
2487
2743
  sourceIds: import_zod12.z.array(import_zod12.z.string()).optional().describe("Capture/restore only these source ids (default all)."),
2488
2744
  deviceId: import_zod12.z.string().optional().describe("Target device id (auto-selected if one is connected).")
2489
2745
  }
@@ -2532,7 +2788,8 @@ ${lines.join("\n")}`);
2532
2788
  const outcome = await call("restore", {
2533
2789
  id: args.id,
2534
2790
  mode: args.mode ?? "live",
2535
- excludeKeys: args.excludeKeys
2791
+ excludeKeys: args.excludeKeys,
2792
+ restoreRoute: args.restoreRoute
2536
2793
  });
2537
2794
  return text11(formatOutcome(outcome));
2538
2795
  }
@@ -2750,11 +3007,14 @@ function formatElement(e) {
2750
3007
  if (e.nativeTag != null) line += ` #${e.nativeTag}`;
2751
3008
  return line;
2752
3009
  }
3010
+ var tapCount = 0;
3011
+ var FLOW_TIP_EVERY = 3;
3012
+ var FLOW_TIP = "\n\nTip: driving several steps? `run_flow` does a whole tap/type/wait/assert sequence in ONE call and reports the final screen plus the network and console output it produced.";
2753
3013
  function registerInteractTools(server, sync) {
2754
3014
  server.registerTool(
2755
3015
  "describe_screen",
2756
3016
  {
2757
- description: "List the interactive / labeled elements currently on screen \u2014 the app's live accessibility-style tree walked from the React fiber tree. Each row shows the component name, visible text/label, role, testID, a normalized tap point (tap: x,y in 0\u20131), and its #nativeTag. Rows marked with \u2022 are tappable (have an onPress). Use this to see what's on screen and pick something to tap with tap_element \u2014 no screenshot needed. Re-run after navigating or when the screen changes, since positions move. Requires Buoy Pro.",
3017
+ description: "List the interactive / labeled elements currently on screen \u2014 the app's live accessibility-style tree walked from the React fiber tree. Each row shows the component name, visible text/label, role, testID, a normalized tap point (tap: x,y in 0\u20131), and its #nativeTag. Rows marked with \u2022 are tappable (have an onPress). Use this to see what's on screen and pick something to tap with tap_element \u2014 no screenshot needed. If you already know the sequence you want to drive, `run_flow` does the whole thing (taps, waits, assertions) in one call instead. Re-run after navigating or when the screen changes, since positions move. Requires Buoy Pro.",
2758
3018
  inputSchema: {
2759
3019
  interactiveOnly: import_zod14.z.boolean().optional().describe("Only return tappable (onPress) elements. Default false (include labels/text for context)."),
2760
3020
  deviceId: import_zod14.z.string().optional().describe("Target device id (auto-selected if one is connected).")
@@ -2783,7 +3043,7 @@ ${elements.map(formatElement).join("\n")}`);
2783
3043
  server.registerTool(
2784
3044
  "tap_element",
2785
3045
  {
2786
- description: "Interact with an on-screen element by invoking its handler directly in JS (no screenshot, no coordinate math, works on real devices). Handles buttons (onPress + onLongPress), toggles/switches and sliders (onValueChange), and text inputs (onChangeText) \u2014 see each element's `control` in describe_screen. Identify the target with an exact `nativeTag` or `testID`, or a fuzzy `query`. Pass `longPress:true` to invoke onLongPress instead of onPress (elements that support it show {long-press} in describe_screen). For a toggle, pass `value:true/false` (omit to flip); for a slider, pass a numeric `value` (omit to jump to the far end); for a text input, pass `text`. If the element is off-screen it's scrolled into view first (scrollIntoView:false to skip). Call describe_screen first to see options + current values. Requires Buoy Pro.",
3046
+ description: "Interact with an on-screen element by invoking its handler directly in JS (no screenshot, no coordinate math, works on real devices). Handles buttons (onPress + onLongPress), toggles/switches and sliders (onValueChange), and text inputs (onChangeText) \u2014 see each element's `control` in describe_screen. Identify the target with an exact `nativeTag` or `testID`, or a fuzzy `query`. Pass `longPress:true` to invoke onLongPress instead of onPress (elements that support it show {long-press} in describe_screen). For a toggle, pass `value:true/false` (omit to flip); for a slider, pass a numeric `value` (omit to jump to the far end); for a text input, pass `text`. If the element is off-screen it's scrolled into view first (scrollIntoView:false to skip). Call describe_screen first to see options + current values. Driving MORE THAN ONE step (reproducing a bug, verifying a fix)? Use `run_flow` instead \u2014 it runs the whole sequence in one call. Requires Buoy Pro.",
2787
3047
  inputSchema: {
2788
3048
  nativeTag: import_zod14.z.number().optional().describe("Exact nativeTag from describe_screen (most precise)."),
2789
3049
  testID: import_zod14.z.string().optional().describe("Exact testID of the element."),
@@ -2815,13 +3075,15 @@ ${elements.map(formatElement).join("\n")}`);
2815
3075
  },
2816
3076
  target
2817
3077
  );
3078
+ tapCount += 1;
3079
+ const flowTip = tapCount >= FLOW_TIP_EVERY && tapCount % FLOW_TIP_EVERY === 0 ? FLOW_TIP : "";
2818
3080
  if (res?.tapped && res.matched) {
2819
3081
  const m = res.matched;
2820
3082
  const label = m.text ?? m.label ?? m.name;
2821
3083
  const id = m.testID ? ` [testID=${m.testID}]` : m.nativeTag != null ? ` #${m.nativeTag}` : "";
2822
3084
  const scrolledNote = res.scrolled ? " (scrolled into view)" : "";
2823
3085
  const verb = m.via === "onValueChange" ? `Set "${label}"${id} \u2192 ${JSON.stringify(m.value)}` : m.via === "onChangeText" ? `Typed ${JSON.stringify(m.value)} into "${label}"${id}` : m.via === "onLongPress" ? `Long-pressed "${label}"${id}` : `Tapped "${label}"${id}`;
2824
- return text12(`${verb}${scrolledNote}.`);
3086
+ return text12(`${verb}${scrolledNote}.${flowTip}`);
2825
3087
  }
2826
3088
  let msg = res?.reason ?? "Tap failed.";
2827
3089
  if (res?.candidates?.length) {
@@ -2909,7 +3171,7 @@ function registerJsThreadTools(server, sync) {
2909
3171
  server.registerTool(
2910
3172
  "get_js_thread_top",
2911
3173
  {
2912
- description: "Task Manager for the React Native JS thread: sample for a few seconds and return a ranked table of which task origins (timers, Promise chains, rAF callbacks, bridge call-ins) consumed JS-thread time \u2014 plus thread busy% and freeze (50ms+ block) attribution. Use when JS FPS is low and you need to know WHAT is eating the thread. To profile one specific interaction, pass clearFirst:true, then drive the UI with tap_element while the sample runs. Drill into a row with get_js_thread_origin_detail. Requires @buoy-gg/js-top on the device.",
3174
+ description: 'Task Manager for the React Native JS thread: sample for a few seconds and return a ranked table of which task origins (timers, Promise chains, rAF callbacks, bridge call-ins) consumed JS-thread time \u2014 plus thread busy% and freeze (50ms+ block) attribution. Use when JS FPS is low and you need to know WHAT is eating the thread. To profile one specific interaction, pass clearFirst:true, then drive the UI with tap_element while the sample runs. Drill into a row with get_js_thread_origin_detail. NOT the tool for React components re-rendering too often \u2014 this measures TASKS on the thread, not renders; for "the page re-renders constantly / typing lags / tapping one button re-renders everything" use measure_renders. Requires @buoy-gg/js-top on the device.',
2913
3175
  inputSchema: {
2914
3176
  deviceId: import_zod15.z.string().optional().describe("Target device id (auto-selected if one is connected)."),
2915
3177
  durationMs: import_zod15.z.number().optional().describe("How long to sample before reading the table. Default 3000, max 30000."),
@@ -3489,6 +3751,62 @@ ${JSON.stringify(slim)}
3489
3751
 
3490
3752
  // src/tools/packs/reload.ts
3491
3753
  var import_zod18 = require("zod");
3754
+
3755
+ // src/capture/devServerReload.ts
3756
+ var import_ws = require("ws");
3757
+ var CONNECT_TIMEOUT_MS = 3e3;
3758
+ function messageSocketUrl(origin) {
3759
+ return `${origin.replace(/^http/, "ws").replace(/\/+$/, "")}/message`;
3760
+ }
3761
+ function reloadViaDevServer(origin) {
3762
+ return new Promise((resolve) => {
3763
+ let settled = false;
3764
+ const done = (result) => {
3765
+ if (settled) return;
3766
+ settled = true;
3767
+ resolve(result);
3768
+ };
3769
+ let socket;
3770
+ try {
3771
+ socket = new import_ws.WebSocket(messageSocketUrl(origin));
3772
+ } catch (e) {
3773
+ done({ ok: false, error: e.message });
3774
+ return;
3775
+ }
3776
+ const timer = setTimeout(() => {
3777
+ try {
3778
+ socket.close();
3779
+ } catch {
3780
+ }
3781
+ done({ ok: false, error: `dev server at ${origin} did not accept a connection` });
3782
+ }, CONNECT_TIMEOUT_MS);
3783
+ let sent = false;
3784
+ socket.on("open", () => {
3785
+ try {
3786
+ socket.send(JSON.stringify({ version: 2, method: "reload" }));
3787
+ sent = true;
3788
+ clearTimeout(timer);
3789
+ setTimeout(() => {
3790
+ try {
3791
+ socket.close();
3792
+ } catch {
3793
+ }
3794
+ done({ ok: true });
3795
+ }, 250);
3796
+ } catch (e) {
3797
+ clearTimeout(timer);
3798
+ done({ ok: false, error: e.message });
3799
+ }
3800
+ });
3801
+ socket.on("error", (e) => {
3802
+ clearTimeout(timer);
3803
+ if (sent) return;
3804
+ done({ ok: false, error: e.message });
3805
+ });
3806
+ });
3807
+ }
3808
+
3809
+ // src/tools/packs/reload.ts
3492
3810
  var APP_TOOL_ID = "app";
3493
3811
  var RELOAD_ACTION = "reloadApp";
3494
3812
  var DEFAULT_WAIT_MS = 3e4;
@@ -3501,7 +3819,11 @@ var errMsg18 = (e) => e instanceof Error ? e.message : String(e);
3501
3819
  var delay3 = (ms) => new Promise((r) => setTimeout(r, ms));
3502
3820
  async function ping(sync, deviceId, timeoutMs) {
3503
3821
  try {
3504
- const pong = await sync.callAction(APP_TOOL_ID, "ping", {}, deviceId, { timeoutMs });
3822
+ const pong = await sync.callAction(APP_TOOL_ID, "ping", {}, deviceId, {
3823
+ timeoutMs,
3824
+ silent: true
3825
+ });
3826
+ sync.noteDevServerOrigin(deviceId, pong?.devServerOrigin);
3505
3827
  return typeof pong?.startedAt === "number" ? pong.startedAt : null;
3506
3828
  } catch {
3507
3829
  return null;
@@ -3530,7 +3852,7 @@ function registerReloadTools(server, sync) {
3530
3852
  server.registerTool(
3531
3853
  "reload_app",
3532
3854
  {
3533
- description: "Reload the running app's JS bundle (Expo Go, dev client, or RN CLI) \u2014 the equivalent of shaking the device and hitting Reload, but from here. Use it to pick up a change Fast Refresh didn't apply, to clear leaked in-memory state before a measurement, or to re-run app startup. In dev builds it uses React Native's DevSettings.reload(); otherwise it falls back to expo-updates (if the app installs it). All in-memory state is lost. By default this waits for the app to come back up and reports how long the reload took. Works on any app running @buoy-gg/core \u2014 no particular tool package needed. Requires Buoy Pro.",
3855
+ description: "Reload the running app's JS bundle (Expo Go, dev client, or RN CLI) \u2014 the equivalent of shaking the device and hitting Reload, but from here. Use it to pick up a change Fast Refresh didn't apply, to clear leaked in-memory state before a measurement, to re-run app startup, or to RECOVER A CRASHED APP: after a fatal error the other tools stop answering, and reloading is how you get them back (it falls through to the dev server when the app is too broken to reload itself). In dev builds it uses React Native's DevSettings.reload(); otherwise it falls back to expo-updates (if the app installs it). All in-memory state is lost \u2014 read anything you need from get_console/get_storage BEFORE reloading. By default this waits for the app to come back up and reports how long the reload took. Works on any app running @buoy-gg/core \u2014 no particular tool package needed. Requires Buoy Pro.",
3534
3856
  inputSchema: {
3535
3857
  strategy: import_zod18.z.enum(["auto", "dev-settings", "expo-updates"]).optional().describe(
3536
3858
  "Reload mechanism. 'auto' (default) uses DevSettings in dev builds and expo-updates otherwise. Force one only when debugging the reload itself."
@@ -3545,13 +3867,27 @@ function registerReloadTools(server, sync) {
3545
3867
  const target = sync.requireProDevice(args.deviceId);
3546
3868
  assertReloadSupported(sync, target);
3547
3869
  const previousRealm = args.wait === false ? null : await ping(sync, target, PROBE_TIMEOUT_MS);
3548
- const ack = await sync.callAction(
3549
- APP_TOOL_ID,
3550
- RELOAD_ACTION,
3551
- { strategy: args.strategy ?? "auto" },
3552
- target
3553
- );
3554
- const via = ack?.strategy ? ` via ${ack.strategy}` : "";
3870
+ let ack = null;
3871
+ let recovered = false;
3872
+ try {
3873
+ ack = await sync.callAction(
3874
+ APP_TOOL_ID,
3875
+ RELOAD_ACTION,
3876
+ { strategy: args.strategy ?? "auto" },
3877
+ target
3878
+ );
3879
+ } catch (inAppError) {
3880
+ const origin = sync.devServerOrigin(target);
3881
+ if (!origin) throw inAppError;
3882
+ const res = await reloadViaDevServer(origin);
3883
+ if (!res.ok) {
3884
+ throw new Error(
3885
+ `The app is not answering (it looks crashed) and the dev server at ${origin} could not be reached either: ${res.error}. Reload the app by hand.`
3886
+ );
3887
+ }
3888
+ recovered = true;
3889
+ }
3890
+ const via = recovered ? ` via the dev server (the app was unresponsive \u2014 its own reload could not be reached)` : ack?.strategy ? ` via ${ack.strategy}` : "";
3555
3891
  if (args.wait === false) {
3556
3892
  return text16(`Reload requested${via}. The app is restarting \u2014 its in-memory state is gone; re-read anything you cached.`);
3557
3893
  }
@@ -3571,12 +3907,836 @@ function registerReloadTools(server, sync) {
3571
3907
  );
3572
3908
  }
3573
3909
 
3910
+ // src/tools/packs/triage.ts
3911
+ var import_zod19 = require("zod");
3912
+ var text17 = (s) => ({ content: [{ type: "text", text: s }] });
3913
+ var fail19 = (message) => ({
3914
+ isError: true,
3915
+ content: [{ type: "text", text: message }]
3916
+ });
3917
+ var errMsg19 = (e) => e instanceof Error ? e.message : String(e);
3918
+ var PROBE_TIMEOUT_MS2 = 5e3;
3919
+ var MAX_STORAGE_KEYS = 20;
3920
+ var MAX_STORAGE_VALUE = 400;
3921
+ var STORAGE_VALUE_BUDGET = 1600;
3922
+ var MAX_LINE = 160;
3923
+ var cap = (s, n = MAX_LINE) => s.length > n ? `${s.slice(0, n)}\u2026` : s;
3924
+ function hhmmss2(ts) {
3925
+ if (!ts) return "--:--:--";
3926
+ const d = new Date(ts);
3927
+ const p = (n) => String(n).padStart(2, "0");
3928
+ return `${p(d.getHours())}:${p(d.getMinutes())}:${p(d.getSeconds())}`;
3929
+ }
3930
+ async function otherBackends(sync, target, opts) {
3931
+ const notes = [];
3932
+ try {
3933
+ const mmkv = await sync.callAction("storage", "mmkv.snapshot", {}, target, opts);
3934
+ for (const inst of mmkv ?? []) {
3935
+ const n = inst.entries?.length ?? 0;
3936
+ if (n > 0) notes.push(`MMKV "${inst.id}": ${n} key${n === 1 ? "" : "s"}`);
3937
+ }
3938
+ } catch {
3939
+ }
3940
+ try {
3941
+ const secure = await sync.callAction("storage", "secure.keys", {}, target, opts);
3942
+ if (secure?.length) notes.push(`SecureStore: ${secure.length} key${secure.length === 1 ? "" : "s"}`);
3943
+ } catch {
3944
+ }
3945
+ return notes.length ? `
3946
+ ${notes.join(" \xB7 ")} \u2014 read with get_storage` : "";
3947
+ }
3948
+ async function section(title, run) {
3949
+ try {
3950
+ const body = await run();
3951
+ return body ? `${title}
3952
+ ${body}` : null;
3953
+ } catch (e) {
3954
+ const msg = errMsg19(e);
3955
+ if (/not responding|timed out|got no answer/i.test(msg)) {
3956
+ return `${title}
3957
+ (no answer \u2014 see the note above)`;
3958
+ }
3959
+ return `${title}
3960
+ (unavailable: ${cap(msg, 120)})`;
3961
+ }
3962
+ }
3963
+ function registerTriageTools(server, sync) {
3964
+ server.registerTool(
3965
+ "get_triage",
3966
+ {
3967
+ description: "START HERE when debugging a running app. One call = the app's current runtime state: connected device, recent console errors (including [FATAL] crashes), recent network requests with statuses, persisted storage values, and what's on screen right now. Use it as the first step for ANY 'why is the app doing X' question \u2014 it is far cheaper than calling get_console + get_network_requests + get_storage + describe_screen separately, and it tells you which of those to drill into next. If the app has crashed it says so and shows the crash. Requires Buoy Pro.",
3968
+ inputSchema: {
3969
+ deviceId: import_zod19.z.string().optional().describe("Target device id (auto-selected if one is connected).")
3970
+ }
3971
+ },
3972
+ async (args) => {
3973
+ try {
3974
+ const target = sync.requireProDevice(args.deviceId);
3975
+ const device = sync.connectedDevices().find((d) => d.deviceId === target);
3976
+ const toolIds = new Set((device?.tools ?? []).map((t) => t.toolId));
3977
+ const head = `**${device?.deviceName ?? target}** (${device?.platform ?? "?"}, id \`${target}\`) \u2014 tools: ${[...toolIds].join(", ") || "none"}`;
3978
+ const opts = { timeoutMs: PROBE_TIMEOUT_MS2 };
3979
+ const [consoleSec, netSec, storeSec, screenSec] = await Promise.all([
3980
+ section("**Console \u2014 recent errors/warnings:**", async () => {
3981
+ if (!toolIds.has("console")) return null;
3982
+ const { payload } = await sync.getSnapshot("console", target, opts);
3983
+ const p = payload;
3984
+ const entries = Array.isArray(p) ? p : p?.entries ?? [];
3985
+ const bad = entries.filter(
3986
+ (e) => e.level === "error" || e.level === "warning"
3987
+ );
3988
+ const shown = bad.slice(-8);
3989
+ if (shown.length === 0) {
3990
+ return `(none \u2014 ${entries.length} entries captured)`;
3991
+ }
3992
+ return shown.map(
3993
+ (e) => `${hhmmss2(e.timestamp)} [${e.level === "error" ? "ERR" : "warn"}] ${cap(
3994
+ (e.message ?? "").split("\n").map((l) => l.trim()).filter(Boolean).slice(0, 3).join(" \u23CE "),
3995
+ 220
3996
+ )}`
3997
+ ).join("\n");
3998
+ }),
3999
+ section("**Network \u2014 recent requests:**", async () => {
4000
+ if (!toolIds.has("network")) return null;
4001
+ const { payload } = await sync.getSnapshot("network", target, opts);
4002
+ const events = Array.isArray(payload) ? payload : payload?.events ?? [];
4003
+ if (events.length === 0) return "(no requests captured yet)";
4004
+ const shown = events.slice(-8);
4005
+ return shown.map((e) => {
4006
+ const mark = e.error || (e.status ?? 0) >= 400 ? "\u274C" : "\u2705";
4007
+ const where = e.path ?? e.url ?? "?";
4008
+ const status = e.error ?? e.status ?? "pending";
4009
+ return `${mark} ${e.method ?? "GET"} ${cap(String(where), 80)} \u2192 ${status}${e.duration ? ` (${Math.round(e.duration)}ms)` : ""}`;
4010
+ }).join("\n");
4011
+ }),
4012
+ section("**Storage \u2014 persisted values:**", async () => {
4013
+ if (!toolIds.has("storage")) return null;
4014
+ const keys = await sync.callAction(
4015
+ "storage",
4016
+ "async.getAllKeys",
4017
+ {},
4018
+ target,
4019
+ opts
4020
+ );
4021
+ if (!Array.isArray(keys) || keys.length === 0) {
4022
+ return "(no app keys)";
4023
+ }
4024
+ const head2 = keys.slice(0, MAX_STORAGE_KEYS);
4025
+ let pairs = [];
4026
+ try {
4027
+ pairs = await sync.callAction(
4028
+ "storage",
4029
+ "async.multiGet",
4030
+ { keys: head2 },
4031
+ target,
4032
+ opts
4033
+ ) ?? [];
4034
+ } catch {
4035
+ return `${head2.map((k) => `\`${k}\``).join(", ")}
4036
+ _Values: get_storage includeValues:true_`;
4037
+ }
4038
+ const byKey = new Map(pairs);
4039
+ let budget = STORAGE_VALUE_BUDGET;
4040
+ const lines = head2.map((k) => {
4041
+ const v = byKey.get(k);
4042
+ if (v == null) return `- \`${k}\` = (null)`;
4043
+ if (v.length > MAX_STORAGE_VALUE) {
4044
+ return `- \`${k}\` = (${(v.length / 1024).toFixed(1)} KB \u2014 read with get_storage)`;
4045
+ }
4046
+ if (budget <= 0) return `- \`${k}\` = (not shown \u2014 triage budget spent; read with get_storage)`;
4047
+ budget -= v.length;
4048
+ return `- \`${k}\` = ${v}`;
4049
+ });
4050
+ const more = keys.length - head2.length;
4051
+ const others = await otherBackends(sync, target, opts);
4052
+ return `${lines.join("\n")}${more > 0 ? `
4053
+ \u2026+${more} more keys` : ""}${others}`;
4054
+ }),
4055
+ section("**Screen \u2014 what's visible now:**", async () => {
4056
+ if (!toolIds.has("highlight-updates")) return null;
4057
+ const res = await sync.callAction(
4058
+ "highlight-updates",
4059
+ "describeScreen",
4060
+ {},
4061
+ target,
4062
+ opts
4063
+ );
4064
+ const els = (res?.elements ?? []).filter((e) => e.interactive);
4065
+ if (els.length === 0) return "(nothing interactive on screen)";
4066
+ return els.slice(0, 20).map(
4067
+ (e) => `\u2022 ${e.name}${e.text || e.label ? ` "${cap(e.text ?? e.label ?? "", 60)}"` : ""}${e.testID ? ` [testID=${e.testID}]` : ""}`
4068
+ ).join("\n");
4069
+ })
4070
+ ]);
4071
+ const crashed = sync.isUnresponsive(target);
4072
+ const banner = crashed ? `
4073
+
4074
+ \u{1F6D1} ${sync.unresponsiveNote(target)}` : "";
4075
+ const body = [consoleSec, netSec, storeSec, screenSec].filter(Boolean).join("\n\n");
4076
+ return text17(
4077
+ `${head}${banner}
4078
+
4079
+ ${body}
4080
+
4081
+ _Drill in with get_console, get_network_requests (includeBodies:true for response bodies), get_storage (includeValues:true), describe_screen._`
4082
+ );
4083
+ } catch (e) {
4084
+ return fail19(errMsg19(e));
4085
+ }
4086
+ }
4087
+ );
4088
+ }
4089
+
4090
+ // src/tools/packs/flow.ts
4091
+ var import_zod20 = require("zod");
4092
+ var UI_TOOL_ID = "highlight-updates";
4093
+ var APP_TOOL_ID2 = "app";
4094
+ var text18 = (s) => ({ content: [{ type: "text", text: s }] });
4095
+ var fail20 = (message) => ({
4096
+ isError: true,
4097
+ content: [{ type: "text", text: message }]
4098
+ });
4099
+ var errMsg20 = (e) => e instanceof Error ? e.message : String(e);
4100
+ var delay4 = (ms) => new Promise((r) => setTimeout(r, ms));
4101
+ var POLL_MS = 250;
4102
+ var DEFAULT_WAIT_MS2 = 5e3;
4103
+ var ASSERT_SETTLE_MS = 2e3;
4104
+ var TEARDOWN_GRACE_MS2 = 1200;
4105
+ var REALM_PROBE_MS = 1500;
4106
+ var MAX_FLOW_MS = 12e4;
4107
+ function matches(e, needle) {
4108
+ if (e.testID && e.testID === needle) return true;
4109
+ const n = needle.toLowerCase();
4110
+ return [e.testID, e.text, e.label, e.name].filter(Boolean).some((v) => String(v).toLowerCase().includes(n));
4111
+ }
4112
+ function formatElement2(e) {
4113
+ const display = e.text ?? e.label;
4114
+ return `${e.interactive ? "\u2022 " : " "}${e.name}${display ? ` "${display}"` : ""}${e.testID ? ` [testID=${e.testID}]` : ""}`;
4115
+ }
4116
+ var stepSchema = import_zod20.z.object({
4117
+ tap: import_zod20.z.string().optional().describe("Tap this element: exact testID, or fuzzy visible text."),
4118
+ longPress: import_zod20.z.boolean().optional().describe("Make `tap` a long-press instead."),
4119
+ type: import_zod20.z.string().optional().describe("Text to type (needs `into`)."),
4120
+ into: import_zod20.z.string().optional().describe("Target text input for `type`: testID or fuzzy text."),
4121
+ set: import_zod20.z.union([import_zod20.z.number(), import_zod20.z.boolean()]).optional().describe("Set a toggle (true/false) or slider (number) \u2014 pair with `tap` as the target."),
4122
+ reload: import_zod20.z.boolean().optional().describe("Reload the JS bundle before this step's action."),
4123
+ waitMs: import_zod20.z.number().min(0).max(3e4).optional().describe("Fixed pause after the action, in ms (max 30000)."),
4124
+ waitFor: import_zod20.z.string().optional().describe("Wait until an element matching this appears (default 5s)."),
4125
+ waitGone: import_zod20.z.string().optional().describe("Wait until no element matches this (default 5s)."),
4126
+ expect: import_zod20.z.string().optional().describe("Assert an element matching this IS on screen (waits up to 2s for it); fails the flow if not."),
4127
+ expectGone: import_zod20.z.string().optional().describe("Assert nothing matches this (waits up to 2s for it to go); fails the flow if something does."),
4128
+ timeoutMs: import_zod20.z.number().min(0).max(6e4).optional().describe("Budget for this step's waitFor/waitGone/expect, max 60000. Default 5000 (2000 for expect).")
4129
+ }).describe("One flow step. Order within a step: reload \u2192 tap/type/set \u2192 waitMs \u2192 waitFor/waitGone \u2192 expect.");
4130
+ function registerFlowTools(server, sync) {
4131
+ server.registerTool(
4132
+ "run_flow",
4133
+ {
4134
+ description: "Run a scripted sequence of UI interactions in ONE call and report what happened: per-step results, the final screen, plus the network requests and console errors the flow produced. Use this instead of alternating describe_screen/tap_element \u2014 reproducing a bug ('open cart, checkout, submit') or verifying a fix after an edit is one call, not a dozen. Steps run in order and stop at the first failure, returning the screen at that point so you can see WHERE it broke. Each step can tap, type, set a toggle/slider, reload, wait for an element, and assert. Requires Buoy Pro.",
4135
+ inputSchema: {
4136
+ steps: import_zod20.z.array(stepSchema).min(1).describe("Steps to run, in order."),
4137
+ include: import_zod20.z.enum(["all", "screen", "none"]).optional().describe("What to report after the flow: 'all' (default) = final screen + network/console produced by the flow; 'screen' = final screen only; 'none' = step results only."),
4138
+ continueOnFailure: import_zod20.z.boolean().optional().describe("Keep going after a failed step instead of stopping. Default false."),
4139
+ deviceId: import_zod20.z.string().optional().describe("Target device id (auto-selected if one is connected).")
4140
+ }
4141
+ },
4142
+ async (args) => {
4143
+ const started = Date.now();
4144
+ try {
4145
+ const target = sync.requireProDevice(args.deviceId);
4146
+ const include = args.include ?? "all";
4147
+ const describe = async () => {
4148
+ const res = await sync.callAction(UI_TOOL_ID, "describeScreen", {}, target);
4149
+ return res?.elements ?? [];
4150
+ };
4151
+ const realmId = async () => {
4152
+ try {
4153
+ const pong = await sync.callAction(APP_TOOL_ID2, "ping", {}, target, {
4154
+ timeoutMs: REALM_PROBE_MS,
4155
+ silent: true
4156
+ // probes across a reload are expected to fail
4157
+ });
4158
+ return typeof pong?.startedAt === "number" ? pong.startedAt : null;
4159
+ } catch {
4160
+ return null;
4161
+ }
4162
+ };
4163
+ const untilNewRealm = async (previous, budgetMs) => {
4164
+ const t0 = Date.now();
4165
+ while (Date.now() - t0 < budgetMs) {
4166
+ const realm = await realmId();
4167
+ if (realm !== null && realm !== previous) return Date.now() - t0;
4168
+ await delay4(POLL_MS);
4169
+ }
4170
+ return null;
4171
+ };
4172
+ const beforeConsole = /* @__PURE__ */ new Set();
4173
+ const beforeNetwork = /* @__PURE__ */ new Set();
4174
+ const toolIds = new Set(
4175
+ (sync.connectedDevices().find((d) => d.deviceId === target)?.tools ?? []).map((t) => t.toolId)
4176
+ );
4177
+ if (include === "all") {
4178
+ if (toolIds.has("console")) {
4179
+ const { payload } = await sync.getSnapshot("console", target).catch(() => ({ payload: null }));
4180
+ const entries = payload?.entries ?? [];
4181
+ for (const e of entries) beforeConsole.add(e.id);
4182
+ }
4183
+ if (toolIds.has("network")) {
4184
+ const { payload } = await sync.getSnapshot("network", target).catch(() => ({ payload: null }));
4185
+ const events = Array.isArray(payload) ? payload : payload?.events ?? [];
4186
+ for (const e of events) beforeNetwork.add(e.id);
4187
+ }
4188
+ }
4189
+ const until = async (predicate, budgetMs) => {
4190
+ const t0 = Date.now();
4191
+ for (; ; ) {
4192
+ if (Date.now() - t0 >= budgetMs || Date.now() - started > MAX_FLOW_MS) {
4193
+ return { ok: false, ms: Date.now() - t0 };
4194
+ }
4195
+ let els = [];
4196
+ try {
4197
+ els = await describe();
4198
+ } catch {
4199
+ }
4200
+ if (predicate(els)) return { ok: true, ms: Date.now() - t0 };
4201
+ await delay4(POLL_MS);
4202
+ }
4203
+ };
4204
+ const lines = [];
4205
+ let failedAt = null;
4206
+ let okCount = 0;
4207
+ for (let i = 0; i < args.steps.length; i++) {
4208
+ const step = args.steps[i];
4209
+ const label = `${i + 1}`;
4210
+ const budget = step.timeoutMs ?? DEFAULT_WAIT_MS2;
4211
+ if (Date.now() - started > MAX_FLOW_MS) {
4212
+ lines.push(`${label} \u2717 flow exceeded ${MAX_FLOW_MS / 1e3}s \u2014 stopped here`);
4213
+ failedAt = i;
4214
+ break;
4215
+ }
4216
+ const stepErrors = [];
4217
+ const done = [];
4218
+ const actionable = step.reload || step.tap || step.type !== void 0 || step.waitMs || step.waitFor || step.waitGone || step.expect || step.expectGone;
4219
+ if (!actionable) {
4220
+ lines.push(`${i + 1} \u2717 step does nothing \u2014 give it a tap/type/wait/expect${step.set !== void 0 ? " (`set` needs `tap` naming the control)" : ""}`);
4221
+ failedAt = failedAt ?? i;
4222
+ if (!args.continueOnFailure) break;
4223
+ continue;
4224
+ }
4225
+ try {
4226
+ if (step.reload) {
4227
+ const previousRealm = await realmId();
4228
+ await sync.callAction(APP_TOOL_ID2, "reloadApp", { strategy: "auto" }, target);
4229
+ await delay4(TEARDOWN_GRACE_MS2);
4230
+ const back = await untilNewRealm(previousRealm, Math.max(budget, 2e4));
4231
+ if (back == null) {
4232
+ stepErrors.push("app did not come back after reload");
4233
+ } else {
4234
+ done.push(`reloaded (${back}ms)`);
4235
+ await until((els) => els.length > 0, 1e4);
4236
+ }
4237
+ }
4238
+ if (step.type !== void 0) {
4239
+ const targetSel = step.into ?? step.tap;
4240
+ if (!targetSel) {
4241
+ stepErrors.push("`type` needs `into` (which input to type into)");
4242
+ } else {
4243
+ const res = await sync.callAction(
4244
+ UI_TOOL_ID,
4245
+ "tapElement",
4246
+ { ...selector(targetSel), text: step.type },
4247
+ target
4248
+ );
4249
+ if (!res?.tapped) {
4250
+ stepErrors.push(res?.reason ?? `could not type into "${targetSel}"`);
4251
+ } else if (res.matched?.via && res.matched.via !== "onChangeText") {
4252
+ stepErrors.push(
4253
+ `"${targetSel}" is not a text input \u2014 it ran ${res.matched.via} instead, and the text was not entered`
4254
+ );
4255
+ } else {
4256
+ done.push(`typed ${JSON.stringify(step.type)} into "${targetSel}"`);
4257
+ }
4258
+ }
4259
+ } else if (step.tap) {
4260
+ const params = { ...selector(step.tap) };
4261
+ if (step.longPress) params.longPress = true;
4262
+ if (step.set !== void 0) params.value = step.set;
4263
+ const res = await sync.callAction(UI_TOOL_ID, "tapElement", params, target);
4264
+ if (!res?.tapped) {
4265
+ stepErrors.push(res?.reason ?? `no element matched "${step.tap}"`);
4266
+ } else {
4267
+ const m = res.matched;
4268
+ const what = m?.text ?? m?.label ?? m?.name ?? step.tap;
4269
+ if (step.set !== void 0 && m?.via && m.via !== "onValueChange") {
4270
+ stepErrors.push(
4271
+ `"${what}" is not a toggle/slider \u2014 it ran ${m.via} instead, so it was not set to ${JSON.stringify(step.set)}`
4272
+ );
4273
+ } else {
4274
+ done.push(
4275
+ step.set !== void 0 ? `set "${what}" \u2192 ${JSON.stringify(m?.value ?? step.set)}` : `${step.longPress ? "long-pressed" : "tapped"} "${what}"`
4276
+ );
4277
+ }
4278
+ }
4279
+ }
4280
+ if (!stepErrors.length && step.waitMs) {
4281
+ await delay4(step.waitMs);
4282
+ done.push(`waited ${step.waitMs}ms`);
4283
+ }
4284
+ if (!stepErrors.length && step.waitFor) {
4285
+ const r = await until((els) => els.some((e) => matches(e, step.waitFor)), budget);
4286
+ if (r.ok) done.push(`"${step.waitFor}" appeared (${r.ms}ms)`);
4287
+ else stepErrors.push(`"${step.waitFor}" never appeared within ${budget}ms`);
4288
+ }
4289
+ if (!stepErrors.length && step.waitGone) {
4290
+ const r = await until((els) => !els.some((e) => matches(e, step.waitGone)), budget);
4291
+ if (r.ok) done.push(`"${step.waitGone}" gone (${r.ms}ms)`);
4292
+ else stepErrors.push(`"${step.waitGone}" was still there after ${budget}ms`);
4293
+ }
4294
+ if (!stepErrors.length && step.expect) {
4295
+ const r = await until(
4296
+ (els) => els.some((e) => matches(e, step.expect)),
4297
+ step.timeoutMs ?? ASSERT_SETTLE_MS
4298
+ );
4299
+ if (r.ok) done.push(`"${step.expect}" is on screen`);
4300
+ else stepErrors.push(`expected "${step.expect}" on screen \u2014 not found`);
4301
+ }
4302
+ if (!stepErrors.length && step.expectGone) {
4303
+ const r = await until(
4304
+ (els) => !els.some((e) => matches(e, step.expectGone)),
4305
+ step.timeoutMs ?? ASSERT_SETTLE_MS
4306
+ );
4307
+ if (r.ok) done.push(`"${step.expectGone}" is gone`);
4308
+ else stepErrors.push(`expected "${step.expectGone}" to be gone \u2014 still on screen`);
4309
+ }
4310
+ } catch (e) {
4311
+ stepErrors.push(errMsg20(e));
4312
+ }
4313
+ if (stepErrors.length) {
4314
+ lines.push(`${label} \u2717 ${describeStep(step)} \u2014 ${stepErrors.join("; ")}`);
4315
+ failedAt = failedAt ?? i;
4316
+ if (!args.continueOnFailure) break;
4317
+ } else {
4318
+ okCount++;
4319
+ lines.push(`${label} \u2713 ${done.join(", ") || describeStep(step)}`);
4320
+ }
4321
+ }
4322
+ const elapsed = ((Date.now() - started) / 1e3).toFixed(1);
4323
+ const header = failedAt === null ? `**Flow: ${okCount}/${args.steps.length} steps OK (${elapsed}s)**` : `**Flow FAILED at step ${failedAt + 1} \u2014 ${okCount}/${args.steps.length} steps OK (${elapsed}s)**`;
4324
+ const parts = [header, lines.join("\n")];
4325
+ if (include !== "none") {
4326
+ try {
4327
+ const els = await describe();
4328
+ const shown = els.filter((e) => e.interactive).slice(0, 30);
4329
+ parts.push(
4330
+ `**Screen now (${shown.length} interactive of ${els.length}):**
4331
+ ${shown.map(formatElement2).join("\n")}`
4332
+ );
4333
+ } catch (e) {
4334
+ parts.push(`**Screen now:** unavailable \u2014 ${cap2(errMsg20(e))}`);
4335
+ }
4336
+ }
4337
+ if (include === "all") {
4338
+ if (toolIds.has("network")) {
4339
+ const { payload } = await sync.getSnapshot("network", target).catch(() => ({ payload: null }));
4340
+ const events = Array.isArray(payload) ? payload : payload?.events ?? [];
4341
+ const fresh = events.filter((e) => !beforeNetwork.has(e.id)).slice(-15);
4342
+ if (fresh.length) {
4343
+ parts.push(
4344
+ `**Network during this flow (${fresh.length}):**
4345
+ ` + fresh.map(
4346
+ (e) => `${e.error || (e.status ?? 0) >= 400 ? "\u274C" : "\u2705"} ${e.method ?? "GET"} ${e.path ?? e.url ?? "?"} \u2192 ${e.error ?? e.status ?? "pending"}`
4347
+ ).join("\n")
4348
+ );
4349
+ }
4350
+ }
4351
+ if (toolIds.has("console")) {
4352
+ const { payload } = await sync.getSnapshot("console", target).catch(() => ({ payload: null }));
4353
+ const entries = payload?.entries ?? [];
4354
+ const fresh = entries.filter((e) => !beforeConsole.has(e.id) && (e.level === "error" || e.level === "warning")).slice(-10);
4355
+ if (fresh.length) {
4356
+ parts.push(
4357
+ `**Console during this flow (${fresh.length}):**
4358
+ ` + fresh.map((e) => `[${e.level === "error" ? "ERR" : "warn"}] ${cap2(firstLines(e.message ?? ""), 220)}`).join("\n")
4359
+ );
4360
+ }
4361
+ }
4362
+ }
4363
+ return text18(parts.join("\n\n"));
4364
+ } catch (e) {
4365
+ return fail20(errMsg20(e));
4366
+ }
4367
+ }
4368
+ );
4369
+ }
4370
+ function selector(needle) {
4371
+ return /\s/.test(needle) ? { query: needle } : { testID: needle, query: needle };
4372
+ }
4373
+ function describeStep(step) {
4374
+ if (step.tap) return `tap "${step.tap}"`;
4375
+ if (step.type !== void 0) return `type into "${step.into ?? "?"}"`;
4376
+ if (step.reload) return "reload";
4377
+ if (step.waitFor) return `waitFor "${step.waitFor}"`;
4378
+ if (step.waitGone) return `waitGone "${step.waitGone}"`;
4379
+ if (step.expect) return `expect "${step.expect}"`;
4380
+ return "step";
4381
+ }
4382
+ var cap2 = (s, n = 160) => s.length > n ? `${s.slice(0, n)}\u2026` : s;
4383
+ var firstLines = (s, n = 3) => s.split("\n").map((l) => l.trim()).filter(Boolean).slice(0, n).join(" \u23CE ");
4384
+
4385
+ // src/tools/packs/renders.ts
4386
+ var import_zod21 = require("zod");
4387
+ var UI_TOOL_ID2 = "highlight-updates";
4388
+ var text19 = (s) => ({ content: [{ type: "text", text: s }] });
4389
+ var fail21 = (message) => ({
4390
+ isError: true,
4391
+ content: [{ type: "text", text: message }]
4392
+ });
4393
+ var errMsg21 = (e) => e instanceof Error ? e.message : String(e);
4394
+ var delay5 = (ms) => new Promise((r) => setTimeout(r, ms));
4395
+ var DEFAULT_IDLE_MS = 3e3;
4396
+ var SETTLE_MS = 400;
4397
+ var DEFAULT_TOP = 12;
4398
+ var lastCapture = /* @__PURE__ */ new Map();
4399
+ var stepSchema2 = import_zod21.z.object({
4400
+ tap: import_zod21.z.string().optional().describe("Tap this element: exact testID, or fuzzy visible text."),
4401
+ type: import_zod21.z.string().optional().describe("Text to type (needs `into`)."),
4402
+ into: import_zod21.z.string().optional().describe("Target text input for `type`: testID or fuzzy text."),
4403
+ waitMs: import_zod21.z.number().min(0).max(1e4).optional().describe("Pause after the action, in ms."),
4404
+ repeat: import_zod21.z.number().min(1).max(20).optional().describe("Do this step N times (default 1). Useful for 'tap it five times'.")
4405
+ }).describe("One interaction to measure. Keep it to the thing you're asking about.");
4406
+ function selector2(needle) {
4407
+ return /\s/.test(needle) ? { query: needle } : { testID: needle, query: needle };
4408
+ }
4409
+ function dominant(causes) {
4410
+ if (!causes) return "unknown";
4411
+ let best = "unknown";
4412
+ let n = -1;
4413
+ for (const [k, v] of Object.entries(causes)) {
4414
+ if ((v ?? 0) > n) {
4415
+ n = v ?? 0;
4416
+ best = k;
4417
+ }
4418
+ }
4419
+ return best;
4420
+ }
4421
+ var CAUSE_HINT = {
4422
+ propsUnstable: "props changed by IDENTITY only (a fresh function/object from the parent) \u2014 fix upstream, not in the child",
4423
+ parent: "parent re-rendered and nothing of its own changed \u2014 a pure cascade",
4424
+ props: "props genuinely changed value",
4425
+ hooks: "its own state/hooks changed",
4426
+ mount: "first render"
4427
+ };
4428
+ function formatCapture(c, top) {
4429
+ const s = c.summary;
4430
+ const wasted = s.wastedRenders ?? 0;
4431
+ const pct = s.totalRenders > 0 ? Math.round(wasted / s.totalRenders * 100) : 0;
4432
+ const lines = [];
4433
+ const ms = s.totalRenderMs ?? 0;
4434
+ lines.push(
4435
+ `**${s.totalRenders} component renders across ${s.totalCommits} commits in ${(c.windowMs / 1e3).toFixed(1)}s**` + (ms > 0 ? ` \xB7 ${ms.toFixed(1)}ms of render work` : "") + ` \xB7 ${wasted} wasted (${pct}%)`
4436
+ );
4437
+ const causeLine = Object.entries(s.causes ?? {}).filter(([, v]) => (v ?? 0) > 0).sort((a, b) => (b[1] ?? 0) - (a[1] ?? 0)).map(([k, v]) => `${k} ${v}`).join(" \xB7 ");
4438
+ if (causeLine) lines.push(`causes: ${causeLine}`);
4439
+ lines.push("");
4440
+ lines.push("| component | self ms | renders | cause | dragged along by | changed props |");
4441
+ lines.push("|---|---:|---:|---|---|---|");
4442
+ const rows = [...s.topComponents].sort((a, b) => (b.totalMs ?? 0) - (a.totalMs ?? 0) || b.renders - a.renders).slice(0, top);
4443
+ for (const r of rows) {
4444
+ lines.push(
4445
+ `| ${r.name} | ${(r.totalMs ?? 0).toFixed(1)} | ${r.renders} | ${dominant(r.causes)} | ${r.parentName ?? "\u2014"} | ${(r.changedProps ?? []).join(", ") || "\u2014"} |`
4446
+ );
4447
+ }
4448
+ if (s.topComponents.length > top) {
4449
+ lines.push(`_\u2026+${s.topComponents.length - top} more components_`);
4450
+ }
4451
+ const seen = new Set(rows.map((r) => dominant(r.causes)));
4452
+ const hints = [...seen].filter((k) => CAUSE_HINT[k]).map((k) => `\`${k}\` = ${CAUSE_HINT[k]}`);
4453
+ if (hints.length) {
4454
+ lines.push("");
4455
+ lines.push(hints.join(" \xB7 "));
4456
+ }
4457
+ lines.push("");
4458
+ lines.push(
4459
+ "`wasted` = renders that produced nothing new (parent cascades + identity-only prop changes) \u2014 the removable part. Sorted by **self ms**: fix what costs time, not what renders often. Cheap components re-rendering is usually fine, and blanket `memo()` trades real code clarity for a number that was never the problem."
4460
+ );
4461
+ return lines.join("\n");
4462
+ }
4463
+ function formatDelta(before, after) {
4464
+ const d = (a, b) => {
4465
+ const diff = b - a;
4466
+ return `${a} \u2192 ${b} (${diff > 0 ? "+" : ""}${diff})`;
4467
+ };
4468
+ const bs = before.summary;
4469
+ const as_ = after.summary;
4470
+ const lines = [`**vs "${before.label}"**`];
4471
+ lines.push(`- total renders: ${d(bs.totalRenders, as_.totalRenders)}`);
4472
+ lines.push(`- wasted: ${d(bs.wastedRenders ?? 0, as_.wastedRenders ?? 0)}`);
4473
+ const beforeByName = new Map(bs.topComponents.map((r) => [r.name, r.renders]));
4474
+ const moved = [];
4475
+ for (const r of as_.topComponents.slice(0, 20)) {
4476
+ const was = beforeByName.get(r.name) ?? 0;
4477
+ if (was !== r.renders) moved.push(`${r.name} ${d(was, r.renders)}`);
4478
+ beforeByName.delete(r.name);
4479
+ }
4480
+ for (const [name, was] of beforeByName) {
4481
+ if (was > 0) moved.push(`${name} ${d(was, 0)}`);
4482
+ }
4483
+ if (moved.length) lines.push(`- per component: ${moved.slice(0, 10).join(" \xB7 ")}`);
4484
+ const fewer = as_.totalRenders < bs.totalRenders;
4485
+ const more = as_.totalRenders > bs.totalRenders;
4486
+ lines.push(
4487
+ more ? "\n\u26A0\uFE0F MORE renders than before \u2014 this change made this interaction worse. Re-render fixes routinely move cost instead of removing it; check the other interactions too before moving on." : fewer ? "\n\u2705 Fewer renders than before on this interaction. Measure the OTHER interactions too \u2014 a fix that helps typing can cost taps." : "\n\u2796 No change on this interaction."
4488
+ );
4489
+ return lines.join("\n");
4490
+ }
4491
+ function registerRenderTools(server, sync) {
4492
+ server.registerTool(
4493
+ "measure_renders",
4494
+ {
4495
+ description: "Find out WHAT re-rendered on the current screen and WHY, in one call \u2014 no test route, no scaffolding. THE tool for 'this page is slow/laggy/janky', 'typing lags', 'the whole page re-renders when I tap one button'. Give it `steps` to measure a specific interaction (tap a button, type in a field) \u2014 measuring an idle window instead tells you nothing about what taps cost, which is how re-render fixes usually go wrong. Returns per-component render COST in ms (sorted by it \u2014 a component rendering 40x for 0.4ms is noise, one rendering 4x for 38ms is the answer), the CAUSE of each (parent cascade / own state / changed props), which parent dragged it along, and which props changed. `wasted` counts renders that produced nothing new \u2014 that's the removable part. AFTER YOU EDIT, call it again with the same steps and compareToPrevious:true to prove the fix landed and didn't move the cost somewhere else. (For choosing between several implementations of a screen, use run_benchmark_batch instead \u2014 that ranks variants; this explains one screen.) Requires Buoy Pro and @buoy-gg/highlight-updates.",
4496
+ inputSchema: {
4497
+ steps: import_zod21.z.array(stepSchema2).max(20).optional().describe(
4498
+ "The interaction to measure, e.g. [{tap:'favorite-deal-2', repeat:5}] or [{type:'bench', into:'deals-search'}]. Omit to sample the screen idle \u2014 only useful for churn that happens with nobody touching it."
4499
+ ),
4500
+ idleMs: import_zod21.z.number().min(500).max(2e4).optional().describe(`Idle sample length when no steps are given. Default ${DEFAULT_IDLE_MS}.`),
4501
+ label: import_zod21.z.string().optional().describe('Name this measurement, e.g. "before fix". Shown in the next comparison.'),
4502
+ compareToPrevious: import_zod21.z.boolean().optional().describe("Diff against the previous measure_renders on this device. Use after an edit to prove the fix."),
4503
+ top: import_zod21.z.number().min(1).max(40).optional().describe(`Components to list. Default ${DEFAULT_TOP}.`),
4504
+ deviceId: import_zod21.z.string().optional().describe("Target device id (auto-selected if one is connected).")
4505
+ }
4506
+ },
4507
+ async (args) => {
4508
+ try {
4509
+ const target = sync.requireProDevice(args.deviceId);
4510
+ const device = sync.connectedDevices().find((d) => d.deviceId === target);
4511
+ const uiTool = (device?.tools ?? []).find((t) => t.toolId === UI_TOOL_ID2);
4512
+ if (!uiTool) {
4513
+ return fail21(
4514
+ "This app doesn't expose the highlight-updates tool, so renders can't be measured. Install @buoy-gg/highlight-updates."
4515
+ );
4516
+ }
4517
+ if (!(uiTool.actions ?? []).includes("beginMeasurement")) {
4518
+ return fail21(
4519
+ "This app's @buoy-gg/highlight-updates is too old to measure renders (needs the version with beginMeasurement). Upgrade it, then try again. Meanwhile run_benchmark_batch with captureRenders can still give counts if you have a test route."
4520
+ );
4521
+ }
4522
+ const began = await sync.callAction(UI_TOOL_ID2, "beginMeasurement", {}, target);
4523
+ if (began && began.ok === false) {
4524
+ return fail21(`Can't measure renders: ${began.reason ?? "capture unavailable"}`);
4525
+ }
4526
+ await delay5(SETTLE_MS);
4527
+ const startedAt = Date.now();
4528
+ const notes = [];
4529
+ if (args.steps?.length) {
4530
+ for (const step of args.steps) {
4531
+ const times = step.repeat ?? 1;
4532
+ for (let i = 0; i < times; i++) {
4533
+ if (step.type !== void 0) {
4534
+ const sel = step.into ?? step.tap;
4535
+ if (!sel) {
4536
+ notes.push("a `type` step needs `into`");
4537
+ continue;
4538
+ }
4539
+ const res = await sync.callAction(
4540
+ UI_TOOL_ID2,
4541
+ "tapElement",
4542
+ { ...selector2(sel), text: step.type },
4543
+ target
4544
+ );
4545
+ if (!res?.tapped) notes.push(res?.reason ?? `could not type into "${sel}"`);
4546
+ else if (res.matched?.via && res.matched.via !== "onChangeText") {
4547
+ notes.push(`"${sel}" is not a text input \u2014 it ran ${res.matched.via}, nothing was typed`);
4548
+ }
4549
+ } else if (step.tap) {
4550
+ const res = await sync.callAction(
4551
+ UI_TOOL_ID2,
4552
+ "tapElement",
4553
+ selector2(step.tap),
4554
+ target
4555
+ );
4556
+ if (!res?.tapped) notes.push(res?.reason ?? `no element matched "${step.tap}"`);
4557
+ }
4558
+ if (step.waitMs) await delay5(step.waitMs);
4559
+ }
4560
+ }
4561
+ await delay5(SETTLE_MS);
4562
+ } else {
4563
+ await delay5(args.idleMs ?? DEFAULT_IDLE_MS);
4564
+ }
4565
+ const windowMs = Date.now() - startedAt;
4566
+ const ended = await sync.callAction(UI_TOOL_ID2, "endMeasurement", {}, target);
4567
+ const summary = ended?.summary;
4568
+ if (!summary) {
4569
+ return fail21(
4570
+ "The capture window closed without returning anything. If the app reloaded mid-measurement, try again."
4571
+ );
4572
+ }
4573
+ const capture = {
4574
+ label: args.label ?? (args.steps?.length ? "interaction" : "idle"),
4575
+ windowMs,
4576
+ summary
4577
+ };
4578
+ const prev = lastCapture.get(target);
4579
+ lastCapture.set(target, capture);
4580
+ const what = args.steps?.length ? args.steps.map((s) => (s.type !== void 0 ? `type "${s.type}"` : `tap "${s.tap}"`) + (s.repeat ? ` \xD7${s.repeat}` : "")).join(", ") : `idle ${(windowMs / 1e3).toFixed(1)}s`;
4581
+ const parts = [`**measure_renders \u2014 ${what}**`, "", formatCapture(capture, args.top ?? DEFAULT_TOP)];
4582
+ if (args.compareToPrevious) {
4583
+ parts.push("", prev ? formatDelta(prev, capture) : "_No previous measurement on this device to compare against._");
4584
+ }
4585
+ if (notes.length) parts.push("", `\u26A0\uFE0F ${notes.join("; ")}`);
4586
+ if (summary.totalRenders === 0) {
4587
+ parts.push(
4588
+ "",
4589
+ "_Nothing re-rendered in this window. If you expected renders, check the app is on the screen you meant and that the steps actually hit something._"
4590
+ );
4591
+ } else if (!args.compareToPrevious) {
4592
+ parts.push("", "_Fixed something? Call this again with the same steps and `compareToPrevious:true` to prove it landed._");
4593
+ }
4594
+ return text19(parts.join("\n"));
4595
+ } catch (e) {
4596
+ return fail21(errMsg21(e));
4597
+ }
4598
+ }
4599
+ );
4600
+ }
4601
+
4602
+ // src/verify/editWatcher.ts
4603
+ var import_node_fs3 = __toESM(require("fs"));
4604
+ var import_node_path3 = __toESM(require("path"));
4605
+ var WATCHED_EXT = /* @__PURE__ */ new Set([
4606
+ ".ts",
4607
+ ".tsx",
4608
+ ".js",
4609
+ ".jsx",
4610
+ ".mjs",
4611
+ ".cjs",
4612
+ ".json"
4613
+ ]);
4614
+ var IGNORED_DIRS = /* @__PURE__ */ new Set([
4615
+ "node_modules",
4616
+ ".git",
4617
+ ".expo",
4618
+ "ios",
4619
+ "android",
4620
+ "dist",
4621
+ "lib",
4622
+ "build",
4623
+ ".next",
4624
+ "coverage",
4625
+ "__snapshots__"
4626
+ ]);
4627
+ var DEBOUNCE_MS = 400;
4628
+ function resolveMode(raw) {
4629
+ const v = (raw ?? "").trim().toLowerCase();
4630
+ if (v === "always" || v === "never" || v === "auto") return v;
4631
+ return "auto";
4632
+ }
4633
+ var EditWatcher = class {
4634
+ constructor() {
4635
+ this.watcher = null;
4636
+ this.changed = /* @__PURE__ */ new Set();
4637
+ this.lastEditAt = 0;
4638
+ this.nudgedForCurrentBatch = false;
4639
+ this.mode = "auto";
4640
+ this.started = false;
4641
+ }
4642
+ start(rootDir, mode) {
4643
+ this.mode = mode;
4644
+ if (mode === "never" || this.started) return;
4645
+ this.started = true;
4646
+ try {
4647
+ this.watcher = import_node_fs3.default.watch(
4648
+ rootDir,
4649
+ { recursive: true },
4650
+ (_event, filename) => {
4651
+ if (!filename) return;
4652
+ const rel2 = String(filename);
4653
+ if (rel2.split(import_node_path3.default.sep).some((seg) => IGNORED_DIRS.has(seg))) return;
4654
+ if (!WATCHED_EXT.has(import_node_path3.default.extname(rel2))) return;
4655
+ this.noteEdit(rel2);
4656
+ }
4657
+ );
4658
+ this.watcher.on("error", () => this.stop());
4659
+ } catch {
4660
+ this.stop();
4661
+ }
4662
+ }
4663
+ /** Record an edit. Public so the nudge policy can be tested without fs.watch. */
4664
+ noteEdit(rel2) {
4665
+ const now = Date.now();
4666
+ if (now - this.lastEditAt > DEBOUNCE_MS) this.nudgedForCurrentBatch = false;
4667
+ this.lastEditAt = now;
4668
+ if (this.changed.size < 50) this.changed.add(rel2);
4669
+ }
4670
+ stop() {
4671
+ try {
4672
+ this.watcher?.close();
4673
+ } catch {
4674
+ }
4675
+ this.watcher = null;
4676
+ }
4677
+ /** Something on the device was actually checked — the edits are accounted for. */
4678
+ markVerified() {
4679
+ this.changed.clear();
4680
+ this.nudgedForCurrentBatch = false;
4681
+ }
4682
+ hasUnverifiedEdits() {
4683
+ return this.changed.size > 0;
4684
+ }
4685
+ /**
4686
+ * The line to append to a tool result, or null to stay quiet.
4687
+ *
4688
+ * `auto` says it once per burst of edits, so it lands when the agent is
4689
+ * mid-loop and then gets out of the way. `always` repeats while anything is
4690
+ * unverified, for people who want the belt and braces.
4691
+ */
4692
+ nudge() {
4693
+ if (this.mode === "never" || !this.hasUnverifiedEdits()) return null;
4694
+ if (this.mode === "auto") {
4695
+ if (this.nudgedForCurrentBatch) return null;
4696
+ this.nudgedForCurrentBatch = true;
4697
+ }
4698
+ const files = [...this.changed];
4699
+ const shown = files.slice(0, 3).join(", ");
4700
+ const more = files.length > 3 ? ` +${files.length - 3} more` : "";
4701
+ return `
4702
+
4703
+ ---
4704
+ \u26A0\uFE0F **${files.length} source file${files.length === 1 ? "" : "s"} changed since anything was last checked on the device** (${shown}${more}).
4705
+ Before reporting this as fixed, observe it working: re-run the same interaction with \`measure_renders\` (\`compareToPrevious: true\`) or \`run_flow\`, and say what the numbers did. A fix nobody watched work is a guess \u2014 and re-render fixes in particular tend to move cost somewhere else rather than remove it.`;
4706
+ }
4707
+ };
4708
+ var editWatcher = new EditWatcher();
4709
+
3574
4710
  // src/server.ts
3575
4711
  var SERVER_INSTRUCTIONS = `Buoy MCP \u2014 drive Buoy dev tools running in a React Native app from your editor.
3576
4712
 
3577
4713
  Buoy tools live inside the app and sync to a local broker; this server connects to that
3578
- broker as a dashboard client. Start by calling \`list_devices\` to see which device(s) are
3579
- connected and which tools (e.g. "perf-monitor", "network", "react-query") they expose.
4714
+ broker as a dashboard client.
4715
+
4716
+ DEBUGGING A RUNNING APP? Call \`get_triage\` FIRST \u2014 one call returns the connected device,
4717
+ recent console errors (including [FATAL] crashes), recent network requests with statuses,
4718
+ persisted storage keys, and what's on screen. It replaces the usual five-call opening
4719
+ sequence and tells you which tool to drill into. Read runtime facts before hypothesizing
4720
+ from source: the app's actual state beats a guess about what the code should do.
4721
+
4722
+ Routing after triage \u2014 match the SYMPTOM to the tool:
4723
+ - Value wrong/stale/NaN, or broken only after an upgrade \u2192 \`get_storage includeValues:true\`
4724
+ (a bad persisted value is a top root cause, and it survives reinstalls of the JS).
4725
+ - Failed request, wrong data from the API, auth/session bug \u2192 \`get_network_requests
4726
+ includeBodies:true\` and read the response BODY.
4727
+ - Crash or redbox \u2192 \`get_console\` (fatals are captured, and still readable after the app
4728
+ dies), then \`get_storage\` for the state that triggered it.
4729
+ - Data on screen doesn't match the API \u2192 \`get_react_query\` / \`get_zustand_state\` /
4730
+ \`get_redux_state\` / \`get_jotai_state\` for the cache or store between them.
4731
+
4732
+ AFTER YOU EDIT, PROVE IT on the device \u2014 the most common way a fix fails is being incomplete,
4733
+ not being wrong. Reload (\`run_flow\` step \`{reload:true}\`, or \`reload_app\`) and re-drive EVERY
4734
+ screen the symptom could appear on, asserting the corrected value: a price shown in three places
4735
+ has to be right in all three, and a migration only counts if it survives a restart. A fix you
4736
+ haven't watched work is a guess.
4737
+
4738
+ Use \`list_devices\` when you need the full tool/action inventory or several devices are
4739
+ connected (\`get_triage\` already names the active one).
3580
4740
 
3581
4741
  To inspect what's HAPPENING at runtime \u2014 network requests, state changes (redux/zustand/jotai),
3582
4742
  react-query updates, storage writes, route changes, renders \u2014 use \`get_events\` with a \`sources\`
@@ -3595,15 +4755,28 @@ Need to SEE a component (verify a UI change, check styling/layout)? Use \`screen
3595
4755
  a testID/name \u2014 it locates the live component on-device and returns a cropped screenshot inline.
3596
4756
  iOS Simulator only, and only works when this MCP server runs on the same Mac as the booted Simulator.
3597
4757
 
3598
- Need to DRIVE the UI (tap a button, follow a flow)? Use \`describe_screen\` to list the on-screen
3599
- elements (accessibility-style tree with tap points + testIDs, no screenshot), then \`tap_element\`
3600
- ({ testID | query | nativeTag }) to tap one \u2014 it invokes the element's onPress directly in JS, so
3601
- it's fast and works on physical devices too. Re-run \`describe_screen\` after the screen changes.
4758
+ Need to DRIVE the UI? Reach for \`run_flow\` FIRST: it runs a whole scripted sequence
4759
+ (tap / type / set / reload / waitFor / expect) in ONE call and reports the per-step result, the
4760
+ final screen, and the network + console output the flow produced. Reproducing a bug and verifying
4761
+ a fix are one call each instead of a dozen alternating reads and taps, and a failed step reports
4762
+ the screen at the point it broke. Drop to \`describe_screen\` (list on-screen elements with testIDs
4763
+ and tap points) + \`tap_element\` only for one-off pokes or when you don't yet know what's there.
4764
+
4765
+ Screen slow / laggy / janky / stutters, typing lags, or "the whole page re-renders when I tap
4766
+ one thing"? Start with \`measure_renders\` \u2014 one call, no test route needed. Pass \`steps\` for the
4767
+ interaction you're actually asking about (a tap, some typing); an idle sample says nothing about
4768
+ what a tap costs. It returns per-component render counts, WHY each rendered (parent cascade / own
4769
+ state / changed props), which parent dragged it along, and \`wasted\` \u2014 the renders that produced
4770
+ nothing new. After you edit, call it again with the same steps and \`compareToPrevious:true\`:
4771
+ re-render fixes routinely move the cost somewhere else instead of removing it, and only a second
4772
+ measurement catches that.
3602
4773
 
3603
4774
  JS FPS low / app hitching and you need to know WHAT is eating the JS thread? Use
3604
4775
  \`get_js_thread_top\` (a live "Task Manager" of timers, Promise chains, rAF callbacks and
3605
4776
  freezes, ranked by thread time), then \`get_js_thread_origin_detail\` to drill into a row.
3606
4777
  Pass clearFirst:true and drive the UI with \`tap_element\` mid-sample to profile one interaction.
4778
+ This is about TASKS on the thread \u2014 for React components re-rendering too often, it's
4779
+ \`measure_renders\`, not this.
3607
4780
 
3608
4781
  Debugging images (not loading, slow, stale cache, memory)? \`get_images\` lists every RN <Image> /
3609
4782
  expo-image load with cache verdicts and oversize audits; \`image_action\` reloads, retries, or
@@ -3639,11 +4812,46 @@ If no device is connected, open the app with Buoy devtools running and try again
3639
4812
  The Buoy MCP is a Pro feature: \`list_devices\` is always available, but data/action tools
3640
4813
  require the connected app to have an active Buoy Pro license. If they error with "Buoy Pro
3641
4814
  required", tell the user to upgrade to Buoy Pro in the app \u2014 don't try to work around it.`;
4815
+ var VERIFYING_TOOLS = /* @__PURE__ */ new Set([
4816
+ "measure_renders",
4817
+ "run_flow",
4818
+ "run_benchmark_batch",
4819
+ "get_triage",
4820
+ "tap_element",
4821
+ "describe_screen",
4822
+ "get_console",
4823
+ "get_network_requests",
4824
+ "get_events",
4825
+ "get_snapshot"
4826
+ ]);
3642
4827
  function createMcpServer(sync) {
3643
4828
  const server = new import_mcp.McpServer(
3644
4829
  { name: "buoy", version: VERSION },
3645
4830
  { capabilities: { tools: {} }, instructions: SERVER_INSTRUCTIONS }
3646
4831
  );
4832
+ editWatcher.start(process.cwd(), resolveMode(process.env.BUOY_VERIFY));
4833
+ const originalRegister = server.registerTool.bind(server);
4834
+ server.registerTool = ((name, config, handler) => originalRegister(name, config, (async (...args) => {
4835
+ const result = await handler(...args);
4836
+ const errored = result?.isError === true;
4837
+ if (VERIFYING_TOOLS.has(name) && !errored) {
4838
+ editWatcher.markVerified();
4839
+ return result;
4840
+ }
4841
+ const note = editWatcher.nudge();
4842
+ if (!note) return result;
4843
+ const r = result;
4844
+ if (Array.isArray(r?.content)) {
4845
+ return {
4846
+ ...r,
4847
+ content: [...r.content, { type: "text", text: note }]
4848
+ };
4849
+ }
4850
+ return result;
4851
+ })));
4852
+ registerTriageTools(server, sync);
4853
+ registerFlowTools(server, sync);
4854
+ registerRenderTools(server, sync);
3647
4855
  registerGenericTools(server, sync);
3648
4856
  registerPerfMonitorTools(server, sync);
3649
4857
  registerEventsTools(server, sync);
@@ -3665,32 +4873,37 @@ function createMcpServer(sync) {
3665
4873
  return server;
3666
4874
  }
3667
4875
  var deviceIdField = {
3668
- deviceId: import_zod19.z.string().optional().describe(
4876
+ deviceId: import_zod22.z.string().optional().describe(
3669
4877
  'Target device id (from list_devices). Omit to auto-select when exactly one device is connected. Use "All" to broadcast.'
3670
4878
  )
3671
4879
  };
3672
4880
 
3673
4881
  // src/init.ts
3674
4882
  var import_node_child_process2 = require("child_process");
3675
- var import_node_fs3 = __toESM(require("fs"));
4883
+ var import_node_fs4 = __toESM(require("fs"));
3676
4884
  var import_node_os3 = __toESM(require("os"));
3677
- var import_node_path3 = __toESM(require("path"));
4885
+ var import_node_path4 = __toESM(require("path"));
4886
+ function defaultEnv() {
4887
+ return { BUOY_VERIFY: "auto" };
4888
+ }
3678
4889
  var PUBLIC_REGISTRY = "https://registry.npmjs.org";
3679
- var PROBE_TIMEOUT_MS2 = 3e3;
4890
+ var PROBE_TIMEOUT_MS3 = 3e3;
3680
4891
  function runInit(opts = {}) {
3681
4892
  const cwd = opts.cwd ?? process.cwd();
3682
- const skillSrcDir = opts.skillSrcDir ?? import_node_path3.default.resolve(__dirname, "../skill");
4893
+ const skillSrcDir = opts.skillSrcDir ?? import_node_path4.default.resolve(__dirname, "../skill");
3683
4894
  const log = opts.log ?? ((m) => process.stdout.write(m + "\n"));
3684
4895
  const entry = resolveEntry(opts, log);
3685
4896
  const done = [];
3686
- upsertMcpConfig(import_node_path3.default.join(cwd, ".mcp.json"), "mcpServers", entry, cwd, done);
3687
- if (import_node_fs3.default.existsSync(import_node_path3.default.join(cwd, ".cursor"))) {
3688
- upsertMcpConfig(import_node_path3.default.join(cwd, ".cursor", "mcp.json"), "mcpServers", entry, cwd, done);
4897
+ upsertMcpConfig(import_node_path4.default.join(cwd, ".mcp.json"), "mcpServers", entry, cwd, done);
4898
+ if (import_node_fs4.default.existsSync(import_node_path4.default.join(cwd, ".cursor"))) {
4899
+ upsertMcpConfig(import_node_path4.default.join(cwd, ".cursor", "mcp.json"), "mcpServers", entry, cwd, done);
3689
4900
  }
3690
- if (import_node_fs3.default.existsSync(import_node_path3.default.join(cwd, ".vscode"))) {
3691
- upsertMcpConfig(import_node_path3.default.join(cwd, ".vscode", "mcp.json"), "servers", entry, cwd, done);
4901
+ if (import_node_fs4.default.existsSync(import_node_path4.default.join(cwd, ".vscode"))) {
4902
+ upsertMcpConfig(import_node_path4.default.join(cwd, ".vscode", "mcp.json"), "servers", entry, cwd, done);
3692
4903
  }
3693
4904
  installSkills(skillSrcDir, cwd, done);
4905
+ appendDebuggingGuide(import_node_path4.default.join(cwd, "CLAUDE.md"), cwd, done);
4906
+ appendDebuggingGuide(import_node_path4.default.join(cwd, "AGENTS.md"), cwd, done);
3694
4907
  log("Buoy MCP installed:");
3695
4908
  for (const line of done) log(" \u2022 " + line);
3696
4909
  log("");
@@ -3710,7 +4923,7 @@ function runInit(opts = {}) {
3710
4923
  }
3711
4924
  function resolveEntry(opts, log) {
3712
4925
  if (opts.cliPath) {
3713
- return { type: "stdio", command: "node", args: [opts.cliPath] };
4926
+ return { type: "stdio", command: "node", args: [opts.cliPath], env: defaultEnv() };
3714
4927
  }
3715
4928
  const mode = opts.mode ?? "auto";
3716
4929
  if (mode === "npx") {
@@ -3729,21 +4942,21 @@ function resolveEntry(opts, log) {
3729
4942
  return localEntry(opts, log);
3730
4943
  }
3731
4944
  function npxEntry() {
3732
- return { type: "stdio", command: "npx", args: ["-y", `${PACKAGE_NAME}@latest`] };
4945
+ return { type: "stdio", command: "npx", args: ["-y", `${PACKAGE_NAME}@latest`], env: defaultEnv() };
3733
4946
  }
3734
4947
  function localEntry(opts, log) {
3735
- const installDir = opts.installDir ?? process.env.BUOY_MCP_HOME ?? import_node_path3.default.join(import_node_os3.default.homedir(), ".buoy-mcp");
4948
+ const installDir = opts.installDir ?? process.env.BUOY_MCP_HOME ?? import_node_path4.default.join(import_node_os3.default.homedir(), ".buoy-mcp");
3736
4949
  const registry = opts.registry ?? PUBLIC_REGISTRY;
3737
4950
  const cliPath = installLocalCopy(installDir, registry, log);
3738
4951
  if (cliPath) {
3739
- return { type: "stdio", command: "node", args: [cliPath] };
4952
+ return { type: "stdio", command: "node", args: [cliPath], env: defaultEnv() };
3740
4953
  }
3741
4954
  log("\u26A0 local install failed \u2014 writing the npx entry instead (see the error above).");
3742
4955
  return npxEntry();
3743
4956
  }
3744
4957
  function installLocalCopy(installDir, registry, log) {
3745
4958
  const spec = `${PACKAGE_NAME}@${VERSION === "0.0.0-dev" ? "latest" : VERSION}`;
3746
- import_node_fs3.default.mkdirSync(installDir, { recursive: true });
4959
+ import_node_fs4.default.mkdirSync(installDir, { recursive: true });
3747
4960
  const npm = process.platform === "win32" ? "npm.cmd" : "npm";
3748
4961
  log(`Installing ${spec} into ${installDir} from ${registry} \u2026`);
3749
4962
  const res = (0, import_node_child_process2.spawnSync)(
@@ -3766,8 +4979,8 @@ function installLocalCopy(installDir, registry, log) {
3766
4979
  log(`\u26A0 npm install exited with code ${res.status ?? "?"}${res.error ? ` (${res.error.message})` : ""}.`);
3767
4980
  return null;
3768
4981
  }
3769
- const cliPath = import_node_path3.default.join(installDir, "node_modules", PACKAGE_NAME, "dist", "cli.js");
3770
- if (!import_node_fs3.default.existsSync(cliPath)) {
4982
+ const cliPath = import_node_path4.default.join(installDir, "node_modules", PACKAGE_NAME, "dist", "cli.js");
4983
+ if (!import_node_fs4.default.existsSync(cliPath)) {
3771
4984
  log(`\u26A0 install finished but ${cliPath} is missing.`);
3772
4985
  return null;
3773
4986
  }
@@ -3789,7 +5002,7 @@ function effectiveRegistry() {
3789
5002
  }
3790
5003
  function npmConfigGet(key) {
3791
5004
  const npm = process.platform === "win32" ? "npm.cmd" : "npm";
3792
- const res = (0, import_node_child_process2.spawnSync)(npm, ["config", "get", key], { encoding: "utf8", timeout: PROBE_TIMEOUT_MS2 });
5005
+ const res = (0, import_node_child_process2.spawnSync)(npm, ["config", "get", key], { encoding: "utf8", timeout: PROBE_TIMEOUT_MS3 });
3793
5006
  if (res.status !== 0 || typeof res.stdout !== "string") return null;
3794
5007
  const val = res.stdout.trim();
3795
5008
  if (!val || val === "undefined" || val === "null") return null;
@@ -3798,8 +5011,8 @@ function npmConfigGet(key) {
3798
5011
  function hostReachable(registry) {
3799
5012
  const done = (0, import_node_child_process2.spawnSync)(
3800
5013
  process.execPath,
3801
- ["-e", REACHABILITY_PROBE, registry, String(PROBE_TIMEOUT_MS2)],
3802
- { encoding: "utf8", timeout: PROBE_TIMEOUT_MS2 + 2e3 }
5014
+ ["-e", REACHABILITY_PROBE, registry, String(PROBE_TIMEOUT_MS3)],
5015
+ { encoding: "utf8", timeout: PROBE_TIMEOUT_MS3 + 2e3 }
3803
5016
  );
3804
5017
  return done.status === 0;
3805
5018
  }
@@ -3819,52 +5032,129 @@ fetch(url, { method: 'HEAD', signal: c.signal })
3819
5032
  `;
3820
5033
  function upsertMcpConfig(file, key, entry, cwd, done) {
3821
5034
  let json3 = {};
3822
- if (import_node_fs3.default.existsSync(file)) {
5035
+ if (import_node_fs4.default.existsSync(file)) {
3823
5036
  try {
3824
- json3 = JSON.parse(import_node_fs3.default.readFileSync(file, "utf8"));
5037
+ json3 = JSON.parse(import_node_fs4.default.readFileSync(file, "utf8"));
3825
5038
  } catch {
3826
- import_node_fs3.default.copyFileSync(file, file + ".bak");
5039
+ import_node_fs4.default.copyFileSync(file, file + ".bak");
3827
5040
  done.push(`existing ${rel(file, cwd)} was invalid JSON \u2014 backed up to ${rel(file, cwd)}.bak`);
3828
5041
  json3 = {};
3829
5042
  }
3830
5043
  } else {
3831
- import_node_fs3.default.mkdirSync(import_node_path3.default.dirname(file), { recursive: true });
5044
+ import_node_fs4.default.mkdirSync(import_node_path4.default.dirname(file), { recursive: true });
3832
5045
  }
3833
5046
  const servers = json3[key] && typeof json3[key] === "object" ? json3[key] : {};
3834
5047
  const existed = "buoy" in servers;
3835
5048
  servers.buoy = entry;
3836
5049
  json3[key] = servers;
3837
- import_node_fs3.default.writeFileSync(file, JSON.stringify(json3, null, 2) + "\n");
5050
+ import_node_fs4.default.writeFileSync(file, JSON.stringify(json3, null, 2) + "\n");
3838
5051
  done.push(`${existed ? "updated" : "added"} \`buoy\` in ${rel(file, cwd)}`);
3839
5052
  }
5053
+ var GUIDE_HEADING = "# Debugging with Buoy";
5054
+ var DEBUGGING_GUIDE = `${GUIDE_HEADING}
5055
+
5056
+ This app's runtime is inspectable while it runs: the Buoy MCP tools read live
5057
+ state straight from the device. Route by symptom before reading source:
5058
+
5059
+ - \`get_triage\` \u2014 START HERE. One call: device, recent console errors and
5060
+ crashes, recent network requests, storage keys, and what's on screen. It
5061
+ usually names the problem, and always names the tool to drill into next.
5062
+ - \`run_flow\` \u2014 reproduce a bug or verify a fix in ONE call: a scripted
5063
+ tap/type/wait/assert sequence that returns the final screen plus the network
5064
+ and console output it produced. Use it instead of alternating
5065
+ \`describe_screen\` and \`tap_element\` a dozen times.
5066
+ - \`get_events\` \u2014 the universal activity timeline (network, state changes,
5067
+ storage writes, routes, renders). Check it BEFORE driving the UI to
5068
+ reproduce a bug \u2014 the failure is usually already recorded.
5069
+ - \`get_network_requests\` \u2014 API data wrong or calls failing \u2192 read the actual
5070
+ request + response bodies instead of guessing from code. Boot-time requests
5071
+ are captured too.
5072
+ - \`get_console\` \u2014 crashes and errors, including FATAL uncaught JS errors
5073
+ (tagged \`[FATAL]\`, with the component that crashed). Readable even after the
5074
+ app dies, so a redboxed app still tells you why.
5075
+ - \`get_storage\` \u2014 data wrong or missing after an app restart \u2192 read the
5076
+ persisted values FIRST.
5077
+ - \`reload_app\` \u2014 recover a crashed app (the other tools stop answering until
5078
+ you do) or pick up a change Fast Refresh missed.
5079
+ - \`get_snapshot\` \u2014 full current state of any one tool when the compact views
5080
+ aren't enough.
5081
+
5082
+ Runtime facts beat guesses \u2014 and after you edit, prove the fix on the device:
5083
+ reload, then re-drive EVERY screen the symptom could show up on and assert the
5084
+ corrected value. Fixes usually fail by being incomplete, not by being wrong.
5085
+
5086
+ ## Don't report a fix you haven't watched work
5087
+
5088
+ Saying "fixed" is a claim about how the app behaves, so check it the same way
5089
+ you'd check anything else about the app \u2014 by looking at it. Re-run the exact
5090
+ interaction that was broken and report what the numbers did:
5091
+
5092
+ - \`measure_renders\` with the same \`steps\` and \`compareToPrevious: true\` for a
5093
+ performance change \u2014 it prints the before/after delta and flags a change that
5094
+ made things worse.
5095
+ - \`run_flow\` re-driving the broken screen for a behaviour change.
5096
+
5097
+ Two things this catches that reading cannot: a fix that is right but incomplete
5098
+ (the value was wrong in three places and you corrected one), and a fix that
5099
+ moved the cost instead of removing it \u2014 the usual failure mode for re-render
5100
+ work, where a page gets faster to type in and slower to tap.
5101
+
5102
+ If the app changed since anything was last checked, Buoy will say so in a tool
5103
+ result. Whether that's worth acting on is your call \u2014 a doc edit isn't worth a
5104
+ device check, a state-management change is. Set \`BUOY_VERIFY\` in the MCP config
5105
+ to \`always\` (repeat until checked) or \`never\` (silent) to change how often it
5106
+ mentions it.
5107
+ `;
5108
+ function appendDebuggingGuide(file, cwd, done) {
5109
+ let existing = "";
5110
+ if (import_node_fs4.default.existsSync(file)) {
5111
+ try {
5112
+ existing = import_node_fs4.default.readFileSync(file, "utf8");
5113
+ } catch {
5114
+ done.push(`\u26A0 could not read ${rel(file, cwd)} \u2014 skipped the debugging guide`);
5115
+ return;
5116
+ }
5117
+ if (existing.includes(GUIDE_HEADING)) {
5118
+ done.push(`${rel(file, cwd)} already has the "${GUIDE_HEADING}" block`);
5119
+ return;
5120
+ }
5121
+ }
5122
+ const separator = existing === "" ? "" : existing.endsWith("\n") ? "\n" : "\n\n";
5123
+ try {
5124
+ import_node_fs4.default.writeFileSync(file, existing + separator + DEBUGGING_GUIDE);
5125
+ done.push(`${existing ? "appended" : "created"} "${GUIDE_HEADING}" block in ${rel(file, cwd)}`);
5126
+ } catch {
5127
+ done.push(`\u26A0 could not write ${rel(file, cwd)} \u2014 skipped the debugging guide`);
5128
+ }
5129
+ }
3840
5130
  function installSkills(skillSrcDir, cwd, done) {
3841
- if (!import_node_fs3.default.existsSync(skillSrcDir)) {
5131
+ if (!import_node_fs4.default.existsSync(skillSrcDir)) {
3842
5132
  done.push("\u26A0 no bundled skills found \u2014 skipped");
3843
5133
  return;
3844
5134
  }
3845
- const skills = import_node_fs3.default.readdirSync(skillSrcDir, { withFileTypes: true }).filter((e) => e.isDirectory() && import_node_fs3.default.existsSync(import_node_path3.default.join(skillSrcDir, e.name, "SKILL.md")));
5135
+ const skills = import_node_fs4.default.readdirSync(skillSrcDir, { withFileTypes: true }).filter((e) => e.isDirectory() && import_node_fs4.default.existsSync(import_node_path4.default.join(skillSrcDir, e.name, "SKILL.md")));
3846
5136
  if (skills.length === 0) {
3847
5137
  done.push("\u26A0 no skills with a SKILL.md found \u2014 skipped");
3848
5138
  return;
3849
5139
  }
3850
5140
  for (const skill of skills) {
3851
- const from = import_node_path3.default.join(skillSrcDir, skill.name);
3852
- const to = import_node_path3.default.join(cwd, ".claude", "skills", skill.name);
5141
+ const from = import_node_path4.default.join(skillSrcDir, skill.name);
5142
+ const to = import_node_path4.default.join(cwd, ".claude", "skills", skill.name);
3853
5143
  copyDir(from, to);
3854
- done.push(`installed the ${skill.name} skill \u2192 ${rel(import_node_path3.default.join(to, "SKILL.md"), cwd)}`);
5144
+ done.push(`installed the ${skill.name} skill \u2192 ${rel(import_node_path4.default.join(to, "SKILL.md"), cwd)}`);
3855
5145
  }
3856
5146
  }
3857
5147
  function copyDir(from, to) {
3858
- import_node_fs3.default.mkdirSync(to, { recursive: true });
3859
- for (const entry of import_node_fs3.default.readdirSync(from, { withFileTypes: true })) {
3860
- const s = import_node_path3.default.join(from, entry.name);
3861
- const d = import_node_path3.default.join(to, entry.name);
5148
+ import_node_fs4.default.mkdirSync(to, { recursive: true });
5149
+ for (const entry of import_node_fs4.default.readdirSync(from, { withFileTypes: true })) {
5150
+ const s = import_node_path4.default.join(from, entry.name);
5151
+ const d = import_node_path4.default.join(to, entry.name);
3862
5152
  if (entry.isDirectory()) copyDir(s, d);
3863
- else import_node_fs3.default.copyFileSync(s, d);
5153
+ else import_node_fs4.default.copyFileSync(s, d);
3864
5154
  }
3865
5155
  }
3866
5156
  function rel(file, cwd) {
3867
- const r = import_node_path3.default.relative(cwd, file);
5157
+ const r = import_node_path4.default.relative(cwd, file);
3868
5158
  return r && !r.startsWith("..") ? r : file;
3869
5159
  }
3870
5160