@boxes-dev/dvb-runtime 1.0.1072 → 1.0.1074

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.
@@ -43,10 +43,12 @@ Ground rules:
43
43
  boxes.dev page where the user can connect Codex, Claude Code, GitHub, and
44
44
  Slack, and it waits until they click Done, which can take several minutes.
45
45
  Read its first lines, then tell the user what the page is for and give
46
- them the link it printed in case the browser didn't open. Continue once it
47
- reports the connected agents. If the user says they're done while it is
48
- still waiting, they finished on the page with sign-in codes because their
49
- browser couldn't reach this computer; stop it and run
46
+ them the link it printed in case the browser didn't open. If their agent
47
+ is already connected to boxes.dev, explain that they don't need to sign in
48
+ again, but still need to click Done to continue setup. Continue once it
49
+ reports the connected agents. Done also finishes the command when the user
50
+ selects **Use a code instead**. With an older CLI, if the user says they
51
+ clicked Done but the command is still waiting, stop it and run
50
52
  `dvb agent-init status`.
51
53
  Over SSH it prints a link for the user to open instead and exits without
52
54
  waiting; continue when the user says they're done.
@@ -55,11 +57,46 @@ Ground rules:
55
57
 
56
58
  ## Phase 2: Start
57
59
 
58
- Run `dvb agent-init start --project <path>` from the project root. This
59
- validates the project, pins its identity, and seeds the manifest with
60
- everything detectable (agent config, repos, untracked-file defaults), plus
61
- the live instance-type catalog and the team's current default. It prints the
62
- manifest path — that file is now the working document.
60
+ First choose the project folder. Most projects are a single repository: set
61
+ up the repository you were opened in, or the one the user names, unless
62
+ evidence shows that the user develops several repositories together. Evidence
63
+ includes:
64
+
65
+ - tracked files in one repository that refer to another by relative path,
66
+ such as a `file:../shared` dependency, a Compose build context, a `go.work`
67
+ entry, or a script that runs `cd ../api`
68
+ - a README or setup doc that tells developers to clone the repositories side
69
+ by side
70
+ - other repositories attached to your current session as additional working
71
+ directories, or a tracked editor workspace file that lists them
72
+ - the user saying that they work across them
73
+
74
+ Repositories that only sit in the same folder are not evidence. A folder of
75
+ unrelated repositories, such as `~/code` or `~/projects`, is a workspace
76
+ rather than a project. Check only the paths your evidence points to, and do
77
+ not survey the parent folder.
78
+
79
+ When you find evidence, ask the user once: name the repositories, show what
80
+ connects them, and recommend an answer. Without evidence, do not ask whether
81
+ the project needs other repositories.
82
+
83
+ A multi-repository project's folder must directly contain each of its
84
+ repositories. If you were opened in a folder of repositories, use the same
85
+ evidence to decide which ones belong together, and ask the user which
86
+ repository to set up when none clearly do. When the choice is one repository,
87
+ start from that repository so it becomes a standard single-repository
88
+ project. If related repositories do not share a parent folder, explain that
89
+ to the user and let them decide whether to move or clone them.
90
+
91
+ Run `dvb agent-init start --project <path>` with the folder you chose, and
92
+ run later `dvb agent-init` commands from that folder. `start` validates the
93
+ project, pins its identity, and seeds the manifest with everything detectable
94
+ (agent config, repos, untracked-file defaults), plus the live instance-type
95
+ catalog and the team's current default. It prints the manifest path — that
96
+ file is now the working document. When the folder is not itself a
97
+ repository, every repository and top-level file inside it starts selected;
98
+ keep only what belongs in the project in `project.repositories` and
99
+ `project.rootEntries`.
63
100
 
64
101
  Read any worktree, project-overlap, or skipped-repository explanations printed by `start`.
65
102
  Explain them to the user during review and confirm the intended project scope.
@@ -139,11 +176,12 @@ destination paths and get the user's choice before selecting outside files.
139
176
  Record unresolved requirements in `setupNotes` for remote setup and validation.
140
177
 
141
178
  Do not silently choose approval-sensitive options. Before `check`, explicitly
142
- confirm runtime agents and imported agent config/history, dirty and untracked
143
- files, databases, instance type, GitHub access, Git SSH/signing, devbox
144
- lifecycle or Template maintenance configuration (including watch paths), and
145
- any instruction that lets the remote setup agent change tracked project files.
146
- It is fine to recommend defaults, but the user must make the choice.
179
+ confirm the repositories in a multi-repository project, runtime agents and
180
+ imported agent config/history, dirty and untracked files, databases,
181
+ instance type, GitHub access, Git SSH/signing, devbox lifecycle or Template
182
+ maintenance configuration (including watch paths), and any instruction that
183
+ lets the remote setup agent change tracked project files. It is fine to
184
+ recommend defaults, but the user must make the choice.
147
185
 
148
186
  Git access may need outside action. For managed GitHub, use check to identify
149
187
  and complete connection, App repository access, or an account update. If
@@ -24,8 +24,10 @@ Rules:
24
24
  ordinary files and folders alongside child repositories. For example,
25
25
  `repositories: [api, web]` and `rootEntries: [package.json, scripts]` will
26
26
  copy those two repos and those two root entries. Removing a repo from this
27
- selection excludes it from access setup and general copying. Older documents
28
- without these fields retain the existing project selection.
27
+ selection excludes it from access setup and general copying. Select several
28
+ repositories only when the user confirmed a multi-repository project (see
29
+ Phase 2 of `guide flow`). Older documents without these fields retain the
30
+ existing project selection.
29
31
  Update startup scripts and maintenance watches when changing the selection;
30
32
  setup will not rewrite your shell commands.
31
33
  - Set `untracked.copy` independently of `localChanges.copyDirty`. Custom
@@ -112,6 +114,9 @@ path (`.` for the root); each watched path is exact and relative to that
112
114
  repository. Maintenance watches always require current boxes.dev GitHub App
113
115
  read access, even when `githubAccess` uses SSH or skips Git transport.
114
116
 
117
+ Set `templateMaintenance.timeoutSeconds` to an integer from 1 through 480
118
+ seconds (at most 8 minutes).
119
+
115
120
  Setup creates the maintenance policy with its schedule off, records a fresh
116
121
  baseline for every watch, and does not run the script. A later watched change
117
122
  or a schedule the user enables in Project settings can start the first run.
package/dist/bin/dvb.mjs CHANGED
@@ -1,6 +1,6 @@
1
1
  #!/usr/bin/env node
2
2
 
3
- !function(){try{var e="undefined"!=typeof window?window:"undefined"!=typeof global?global:"undefined"!=typeof globalThis?globalThis:"undefined"!=typeof self?self:{},n=(new e.Error).stack;n&&(e._sentryDebugIds=e._sentryDebugIds||{},e._sentryDebugIds[n]="e37678ce-fb44-5903-81f8-a25f3b4c4d9c")}catch(e){}}();
3
+ !function(){try{var e="undefined"!=typeof window?window:"undefined"!=typeof global?global:"undefined"!=typeof globalThis?globalThis:"undefined"!=typeof self?self:{},n=(new e.Error).stack;n&&(e._sentryDebugIds=e._sentryDebugIds||{},e._sentryDebugIds[n]="b8e2b3d8-97c2-56a6-8a4f-a3626d87d781")}catch(e){}}();
4
4
  import { createRequire as __dvbCreateRequire } from "node:module";
5
5
  import { dirname as __dvbDirname } from "node:path";
6
6
  import { fileURLToPath as __dvbFileURLToPath } from "node:url";
@@ -69730,9 +69730,9 @@ var init_sentry = __esm({
69730
69730
  uncaughtExceptionMonitorInstalled = false;
69731
69731
  currentDevboxSentryUser = null;
69732
69732
  readBuildMetadata = () => {
69733
- const rawPackageVersion = "1.0.1072";
69734
- const rawGitSha = "38cca56e185da1e977f029154045c5ef6ebcef5a";
69735
- const rawSentryRelease = "boxes-dev-dvb@1.0.1072+38cca56e185da1e977f029154045c5ef6ebcef5a";
69733
+ const rawPackageVersion = "1.0.1074";
69734
+ const rawGitSha = "36b0586d995fab98581795868e96940b207d13b2";
69735
+ const rawSentryRelease = "boxes-dev-dvb@1.0.1074+36b0586d995fab98581795868e96940b207d13b2";
69736
69736
  const packageVersion = typeof rawPackageVersion === "string" ? rawPackageVersion : void 0;
69737
69737
  const gitSha = typeof rawGitSha === "string" ? rawGitSha : void 0;
69738
69738
  const sentryRelease = typeof rawSentryRelease === "string" ? rawSentryRelease : void 0;
@@ -75116,7 +75116,7 @@ var init_packageVersion = __esm({
75116
75116
  return path18.join(process.cwd(), "dvb");
75117
75117
  };
75118
75118
  readEmbeddedPackageVersion = () => {
75119
- const raw = "1.0.1072";
75119
+ const raw = "1.0.1074";
75120
75120
  return trimVersion(raw);
75121
75121
  };
75122
75122
  readNearestPackageMetadata = (basePath) => {
@@ -217640,7 +217640,7 @@ var init_composerDraftSync = __esm({
217640
217640
  });
217641
217641
 
217642
217642
  // src/devbox/agentView/creationControls.ts
217643
- var formatAgentViewCodexSpeed, requireCodexModel, requireClaudeModel, nextValue, resolveCodexDefaultEffort, resolveClaudeDefaultEffort, formatAgentViewCreationReasoning, createDefaultAgentViewExecutionSelection, cycleAgentViewCreationModel, setAgentViewCreationModel, cycleAgentViewCreationReasoning, setAgentViewCreationReasoning, cycleAgentViewCreationSpeed, setAgentViewCreationSpeed, resolveAgentViewCreationExecutionSelection, presentAgentViewCreationExecution, formatAgentViewCreationStatusLine;
217643
+ var formatAgentViewCodexSpeed, requireCodexModel, requireClaudeModel, listSelectionSpeeds, nextValue, resolveCodexDefaultEffort, resolveClaudeDefaultEffort, formatAgentViewCreationReasoning, createDefaultAgentViewExecutionSelection, cycleAgentViewCreationModel, setAgentViewCreationModel, cycleAgentViewCreationReasoning, setAgentViewCreationReasoning, cycleAgentViewCreationSpeed, setAgentViewCreationSpeed, resolveAgentViewCreationExecutionSelection, presentAgentViewCreationExecution, formatAgentViewCreationStatusLine;
217644
217644
  var init_creationControls = __esm({
217645
217645
  "src/devbox/agentView/creationControls.ts"() {
217646
217646
  "use strict";
@@ -217660,6 +217660,7 @@ var init_creationControls = __esm({
217660
217660
  }
217661
217661
  return model;
217662
217662
  };
217663
+ listSelectionSpeeds = (options, selection) => selection.backend === "codex" ? requireCodexModel(options, selection.model).serviceTiers : requireClaudeModel(options, selection.model).serviceTiers;
217663
217664
  nextValue = (values, current) => {
217664
217665
  if (values.length === 0) {
217665
217666
  throw new Error("Managed execution options must not be empty.");
@@ -217716,7 +217717,7 @@ var init_creationControls = __esm({
217716
217717
  backend: "claude",
217717
217718
  model: model.model,
217718
217719
  reasoningEffort: resolveClaudeDefaultEffort(model),
217719
- serviceTier: null
217720
+ serviceTier: "standard"
217720
217721
  };
217721
217722
  };
217722
217723
  cycleAgentViewCreationModel = (options, selection) => {
@@ -217742,7 +217743,7 @@ var init_creationControls = __esm({
217742
217743
  backend: "claude",
217743
217744
  model: model.model,
217744
217745
  reasoningEffort: resolveClaudeDefaultEffort(model),
217745
- serviceTier: null
217746
+ serviceTier: "standard"
217746
217747
  };
217747
217748
  };
217748
217749
  setAgentViewCreationModel = (options, selection, modelId) => {
@@ -217763,7 +217764,7 @@ var init_creationControls = __esm({
217763
217764
  backend: "claude",
217764
217765
  model: model.model,
217765
217766
  reasoningEffort: resolveClaudeDefaultEffort(model),
217766
- serviceTier: null
217767
+ serviceTier: "standard"
217767
217768
  };
217768
217769
  };
217769
217770
  cycleAgentViewCreationReasoning = (options, selection) => {
@@ -217816,26 +217817,26 @@ var init_creationControls = __esm({
217816
217817
  return { ...selection, reasoningEffort };
217817
217818
  };
217818
217819
  cycleAgentViewCreationSpeed = (options, selection) => {
217819
- if (selection.backend !== "codex") {
217820
- return selection;
217821
- }
217822
- const speeds = requireCodexModel(options, selection.model).serviceTiers;
217823
- const index = speeds.indexOf(selection.serviceTier);
217824
- return {
217825
- ...selection,
217826
- serviceTier: speeds[(index + 1) % speeds.length] ?? "standard"
217827
- };
217820
+ const speeds = listSelectionSpeeds(options, selection);
217821
+ const index = speeds.indexOf(selection.serviceTier ?? "standard");
217822
+ return setAgentViewCreationSpeed(
217823
+ options,
217824
+ selection,
217825
+ speeds[(index + 1) % speeds.length] ?? "standard"
217826
+ );
217828
217827
  };
217829
217828
  setAgentViewCreationSpeed = (options, selection, serviceTier) => {
217830
- if (selection.backend !== "codex") {
217831
- throw new Error("Claude does not expose a speed setting.");
217832
- }
217833
- const model = requireCodexModel(options, selection.model);
217834
- if (serviceTier !== "standard" && !model.serviceTiers.includes(serviceTier)) {
217829
+ if (serviceTier !== "standard" && !listSelectionSpeeds(options, selection).includes(serviceTier)) {
217835
217830
  throw new Error(
217836
- `Managed Codex model ${selection.model} does not support ${formatAgentViewCodexSpeed(serviceTier)}.`
217831
+ `Managed ${selection.backend === "codex" ? "Codex" : "Claude"} model ${selection.model} does not support ${formatAgentViewCodexSpeed(serviceTier)}.`
217837
217832
  );
217838
217833
  }
217834
+ if (selection.backend === "codex") {
217835
+ return { ...selection, serviceTier };
217836
+ }
217837
+ if (serviceTier === "ultrafast") {
217838
+ throw new Error("Claude does not offer Ultrafast.");
217839
+ }
217839
217840
  return { ...selection, serviceTier };
217840
217841
  };
217841
217842
  resolveAgentViewCreationExecutionSelection = (options, backend, stored) => {
@@ -217854,7 +217855,11 @@ var init_creationControls = __esm({
217854
217855
  withModel,
217855
217856
  stored.reasoningEffort
217856
217857
  );
217857
- return stored.backend === "codex" ? setAgentViewCreationSpeed(options, withReasoning, stored.serviceTier) : withReasoning;
217858
+ return setAgentViewCreationSpeed(
217859
+ options,
217860
+ withReasoning,
217861
+ stored.serviceTier ?? "standard"
217862
+ );
217858
217863
  } catch {
217859
217864
  return fallback2;
217860
217865
  }
@@ -217865,8 +217870,9 @@ var init_creationControls = __esm({
217865
217870
  agent: selection.backend === "codex" ? "Codex" : "Claude",
217866
217871
  model: model.label,
217867
217872
  reasoning: formatAgentViewCreationReasoning(selection.reasoningEffort),
217868
- speed: selection.backend === "codex" ? formatAgentViewCodexSpeed(selection.serviceTier) : null,
217869
- canCycleSpeed: selection.backend === "codex" && requireCodexModel(options, selection.model).serviceTiers.length > 1
217873
+ // Claude shows speed only on a model with fast mode.
217874
+ speed: selection.backend === "codex" || listSelectionSpeeds(options, selection).length > 1 ? formatAgentViewCodexSpeed(selection.serviceTier ?? "standard") : null,
217875
+ canCycleSpeed: listSelectionSpeeds(options, selection).length > 1
217870
217876
  };
217871
217877
  };
217872
217878
  formatAgentViewCreationStatusLine = (execution) => [
@@ -217906,13 +217912,8 @@ var init_composerSettingsFocus = __esm({
217906
217912
  if (model.reasoningEfforts.length > 1) {
217907
217913
  segments.push("reasoning");
217908
217914
  }
217909
- if (context34.selection.backend === "codex") {
217910
- const codexModel = context34.options.codex.find(
217911
- (candidate) => candidate.model === context34.selection.model
217912
- );
217913
- if (codexModel && codexModel.serviceTiers.length > 1) {
217914
- segments.push("speed");
217915
- }
217915
+ if (model.serviceTiers.length > 1) {
217916
+ segments.push("speed");
217916
217917
  }
217917
217918
  return segments;
217918
217919
  };
@@ -217955,9 +217956,9 @@ var init_composerSettingsFocus = __esm({
217955
217956
  }));
217956
217957
  }
217957
217958
  case "speed":
217958
- return (options.codex.find((candidate) => candidate.model === selection.model)?.serviceTiers ?? ["standard"]).map((serviceTier) => ({
217959
+ return requireModel(context34).serviceTiers.map((serviceTier) => ({
217959
217960
  label: formatAgentViewCodexSpeed(serviceTier),
217960
- current: selection.backend === "codex" && serviceTier === selection.serviceTier,
217961
+ current: serviceTier === (selection.serviceTier ?? "standard"),
217961
217962
  value: { segment: "speed", serviceTier }
217962
217963
  }));
217963
217964
  }
@@ -220483,7 +220484,7 @@ var init_creationComposerController = __esm({
220483
220484
  promptBytes: validation.utf8Bytes,
220484
220485
  model: composer.executionSelection.model,
220485
220486
  effort: composer.executionSelection.reasoningEffort,
220486
- serviceTier: composer.executionSelection.backend === "codex" ? composer.executionSelection.serviceTier : null,
220487
+ serviceTier: composer.executionSelection.serviceTier,
220487
220488
  imageAttachmentCount: preparedAttachments.length
220488
220489
  });
220489
220490
  void this.client.action(kickoffAgentViewThreadStartV1, {
@@ -220572,7 +220573,7 @@ var init_creationComposerController = __esm({
220572
220573
  backend: composer.backend,
220573
220574
  model: composer.executionSelection.model,
220574
220575
  effort: composer.executionSelection.reasoningEffort,
220575
- serviceTier: composer.executionSelection.backend === "codex" ? composer.executionSelection.serviceTier : null,
220576
+ serviceTier: composer.executionSelection.serviceTier,
220576
220577
  imageAttachmentCount: preparedAttachments.length,
220577
220578
  error: error55 instanceof Error ? error55.message : String(error55)
220578
220579
  });
@@ -231822,8 +231823,12 @@ var init_devboxes2 = __esm({
231822
231823
  renderOperationWarnings = (warnings) => (warnings ?? []).map((warning3) => `
231823
231824
  Warning: ${warning3.message}`).join("");
231824
231825
  renderCreate = (data) => {
231826
+ const failure = data.operation.failure;
231825
231827
  const result = data.target ? `${data.target.name} ${data.target.instanceId} ${data.operation.operationId}` : `Devbox creation ${data.operation.status}: ${data.operation.operationId}`;
231826
- return `${result}${renderOperationWarnings(data.operation.warnings)}`;
231828
+ const details = failure ? `
231829
+ ${failure.code}: ${failure.message}` : "";
231830
+ const retry = failure?.retryable ? "\nTo try again, run devboxes create with a new --idempotency-key. Reusing the previous key will return this failed operation." : "";
231831
+ return `${result}${details}${retry}${renderOperationWarnings(data.operation.warnings)}`;
231827
231832
  };
231828
231833
  renderStatus = (data) => {
231829
231834
  const failure = data.operation.failure;
@@ -231867,14 +231872,18 @@ Warning: ${warning3.message}`).join("");
231867
231872
  projectSelector: parsed.projectSelector,
231868
231873
  idempotencyKey: parsed.idempotencyKey
231869
231874
  });
231870
- return parsed.wait ? await dependencies.waitForOperation({
231875
+ const result = parsed.wait ? await dependencies.waitForOperation({
231871
231876
  context: context34,
231872
231877
  creation,
231873
231878
  projectSelector: parsed.projectSelector
231874
231879
  }) : creation;
231880
+ if (parsed.wait && result.operation.status === "failed") {
231881
+ dependencies.setExitCode(1);
231882
+ }
231883
+ return result;
231875
231884
  },
231876
231885
  render: renderCreate,
231877
- nextActions: (data) => data.operation.status === "pending" || data.operation.status === "running" ? [
231886
+ nextActions: (data) => data.operation.status === "pending" || data.operation.status === "running" || data.operation.status === "failed" ? [
231878
231887
  {
231879
231888
  command: buildResourceNextActionCommand([
231880
231889
  "dvb",
@@ -231884,7 +231893,7 @@ Warning: ${warning3.message}`).join("");
231884
231893
  data.operation.operationId,
231885
231894
  "--json"
231886
231895
  ]),
231887
- description: "Read the durable creation result."
231896
+ description: data.operation.failure?.retryable ? "Read the failure details. To try again, run devboxes create with a new --idempotency-key; reusing the previous key will return the same operation." : "Read the durable creation result."
231888
231897
  }
231889
231898
  ] : []
231890
231899
  })
@@ -250251,6 +250260,46 @@ var init_codexSignIn = __esm({
250251
250260
  }
250252
250261
  });
250253
250262
 
250263
+ // src/devbox/agentInit/accountsCompletion.ts
250264
+ var accountsCompletionForViewer, waitForRemoteAccountsCompletion;
250265
+ var init_accountsCompletion = __esm({
250266
+ "src/devbox/agentInit/accountsCompletion.ts"() {
250267
+ "use strict";
250268
+ init_server();
250269
+ accountsCompletionForViewer = makeFunctionReference("agentInitProgress:getAccountsCompletionForViewer");
250270
+ waitForRemoteAccountsCompletion = async ({
250271
+ client,
250272
+ previousCompletion,
250273
+ signal
250274
+ }) => {
250275
+ signal.throwIfAborted();
250276
+ let unsubscribe4 = () => {
250277
+ };
250278
+ let onAbort = () => {
250279
+ };
250280
+ try {
250281
+ await new Promise((resolve3, reject) => {
250282
+ onAbort = () => reject(signal.reason);
250283
+ signal.addEventListener("abort", onAbort, { once: true });
250284
+ unsubscribe4 = client.onUpdate(
250285
+ accountsCompletionForViewer,
250286
+ {},
250287
+ (completedAt) => {
250288
+ if (completedAt !== null && completedAt !== previousCompletion) {
250289
+ resolve3();
250290
+ }
250291
+ },
250292
+ reject
250293
+ );
250294
+ });
250295
+ } finally {
250296
+ signal.removeEventListener("abort", onAbort);
250297
+ unsubscribe4();
250298
+ }
250299
+ };
250300
+ }
250301
+ });
250302
+
250254
250303
  // src/devbox/agentInit/codingAgents.ts
250255
250304
  var AGENT_INIT_CONNECT_AGENTS_ACTION, isCodingAgentConnected, fetchConnectedCodingAgents;
250256
250305
  var init_codingAgents2 = __esm({
@@ -251266,6 +251315,7 @@ var init_accounts = __esm({
251266
251315
  init_accountScope2();
251267
251316
  init_accountsServer();
251268
251317
  init_codexSignIn();
251318
+ init_accountsCompletion();
251269
251319
  init_service();
251270
251320
  init_progressReport();
251271
251321
  init_remoteSession();
@@ -251413,78 +251463,93 @@ var init_accounts = __esm({
251413
251463
  });
251414
251464
  return;
251415
251465
  }
251416
- const codexSignInAbort = new AbortController();
251417
- const handlers2 = createAccountsHandlers({
251418
- accountScope,
251419
- codexSignInAbort,
251420
- controlPlaneToken
251421
- });
251422
- const server = await startAccountsServer({
251423
- buildConnectPageUrl: ({ port, nonce }, pageReturn) => buildAccountsConnectPageUrl({
251466
+ await withConvexClient(controlPlaneToken, async (client) => {
251467
+ const previousCompletion = await client.query(
251468
+ accountsCompletionForViewer,
251469
+ {}
251470
+ );
251471
+ const codexSignInAbort = new AbortController();
251472
+ const handlers2 = createAccountsHandlers({
251473
+ accountScope,
251474
+ codexSignInAbort,
251475
+ controlPlaneToken
251476
+ });
251477
+ const server = await startAccountsServer({
251478
+ buildConnectPageUrl: ({ port, nonce }, pageReturn) => buildAccountsConnectPageUrl({
251479
+ email: viewer.email ?? null,
251480
+ nonce,
251481
+ pageReturn,
251482
+ port,
251483
+ siteUrl,
251484
+ userId: viewer.userId
251485
+ }),
251486
+ handlers: handlers2,
251487
+ idleTimeoutMs: IDLE_TIMEOUT_MS,
251488
+ logError: (step, error55) => {
251489
+ const label = step === "codex" ? "Codex sign-in" : step === "claude" ? "Claude Code sign-in" : "Saving the coding agent choice";
251490
+ console.log(`${label} failed: ${errorMessage(error55)}`);
251491
+ }
251492
+ });
251493
+ const pageUrl = buildAccountsConnectPageUrl({
251424
251494
  email: viewer.email ?? null,
251425
- nonce,
251426
- pageReturn,
251427
- port,
251495
+ nonce: server.nonce,
251496
+ pageReturn: { kind: "page" },
251497
+ port: server.port,
251428
251498
  siteUrl,
251429
251499
  userId: viewer.userId
251430
- }),
251431
- handlers: handlers2,
251432
- idleTimeoutMs: IDLE_TIMEOUT_MS,
251433
- logError: (step, error55) => {
251434
- const label = step === "codex" ? "Codex sign-in" : step === "claude" ? "Claude Code sign-in" : "Saving the coding agent choice";
251435
- console.log(`${label} failed: ${errorMessage(error55)}`);
251500
+ });
251501
+ let finishedBy;
251502
+ try {
251503
+ console.log(
251504
+ openBrowser(pageUrl) ? "Opened the boxes.dev connect page in the browser:" : "Couldn't open the browser. Ask the user to open this link:"
251505
+ );
251506
+ console.log(` ${pageUrl}`);
251507
+ console.log(
251508
+ "Waiting for the user to review their accounts and click Done. They need to click Done even if Codex or Claude Code is already connected."
251509
+ );
251510
+ await reportAgentInitStepBestEffort({
251511
+ controlPlaneToken,
251512
+ step: { command: "accounts", outcome: "waiting", phase: null }
251513
+ });
251514
+ finishedBy = await Promise.race([
251515
+ server.done.then(() => "done"),
251516
+ server.idle.then(() => "idle"),
251517
+ waitForRemoteAccountsCompletion({
251518
+ client,
251519
+ previousCompletion,
251520
+ signal: codexSignInAbort.signal
251521
+ }).then(() => "remote")
251522
+ ]);
251523
+ } finally {
251524
+ codexSignInAbort.abort();
251525
+ await server.close();
251526
+ }
251527
+ if (finishedBy !== "done" && !await handlers2.finishAccounts()) {
251528
+ const message = finishedBy === "idle" ? "Stopped waiting after 20 minutes without activity, and no coding agent is connected yet. Run `dvb agent-init accounts` again when the user is ready." : "No coding agent is connected. Run `dvb agent-init accounts` again to connect Codex or Claude Code.";
251529
+ console.log(message);
251530
+ await reportAgentInitStepBestEffort({
251531
+ controlPlaneToken,
251532
+ step: {
251533
+ command: "accounts",
251534
+ outcome: "failed",
251535
+ phase: null,
251536
+ detail: message
251537
+ }
251538
+ });
251539
+ process.exitCode = 1;
251540
+ return;
251541
+ }
251542
+ const agents = handlers2.connectedAgents();
251543
+ if (!agents) {
251544
+ throw new Error("The coding agent choice was not saved.");
251436
251545
  }
251437
- });
251438
- const pageUrl = buildAccountsConnectPageUrl({
251439
- email: viewer.email ?? null,
251440
- nonce: server.nonce,
251441
- pageReturn: { kind: "page" },
251442
- port: server.port,
251443
- siteUrl,
251444
- userId: viewer.userId
251445
- });
251446
- console.log(
251447
- openBrowser(pageUrl) ? "Opened the boxes.dev connect page in the browser:" : "Couldn't open the browser. Ask the user to open this link:"
251448
- );
251449
- console.log(` ${pageUrl}`);
251450
- console.log(
251451
- "Waiting for the user to connect Codex or Claude Code and click Done. This can take several minutes."
251452
- );
251453
- await reportAgentInitStepBestEffort({
251454
- controlPlaneToken,
251455
- step: { command: "accounts", outcome: "waiting", phase: null }
251456
- });
251457
- const finishedBy = await Promise.race([
251458
- server.done.then(() => "done"),
251459
- server.idle.then(() => "idle")
251460
- ]);
251461
- codexSignInAbort.abort();
251462
- await server.close();
251463
- if (finishedBy === "idle" && !await handlers2.finishAccounts()) {
251464
- const message = "Stopped waiting after 20 minutes without activity, and no coding agent is connected yet. Run `dvb agent-init accounts` again when the user is ready.";
251465
- console.log(message);
251466
251546
  await reportAgentInitStepBestEffort({
251467
251547
  controlPlaneToken,
251468
- step: {
251469
- command: "accounts",
251470
- outcome: "failed",
251471
- phase: null,
251472
- detail: message
251473
- }
251548
+ step: { command: "accounts", outcome: "ok", phase: null }
251474
251549
  });
251475
- process.exitCode = 1;
251476
- return;
251477
- }
251478
- const agents = handlers2.connectedAgents();
251479
- if (!agents) {
251480
- throw new Error("The coding agent choice was not saved.");
251481
- }
251482
- await reportAgentInitStepBestEffort({
251483
- controlPlaneToken,
251484
- step: { command: "accounts", outcome: "ok", phase: null }
251550
+ console.log(`Connected to boxes.dev: ${describeConnectedAgents(agents)}.`);
251551
+ console.log("Run `dvb agent-init status` for the next step.");
251485
251552
  });
251486
- console.log(`Connected to boxes.dev: ${describeConnectedAgents(agents)}.`);
251487
- console.log("Run `dvb agent-init status` for the next step.");
251488
251553
  };
251489
251554
  }
251490
251555
  });
@@ -264404,6 +264469,18 @@ var init_seed = __esm({
264404
264469
  push("");
264405
264470
  push("project:");
264406
264471
  push(` root: ${quote(projectRoot)}`);
264472
+ if (repos.some((repo) => repo.relPath !== ".")) {
264473
+ push(
264474
+ " # TODO(agent): this folder is not a repository, so every repository and"
264475
+ );
264476
+ push(
264477
+ " # top-level file inside it is selected below. Keep only what belongs in"
264478
+ );
264479
+ push(
264480
+ " # the project (see Phase 2 of `dvb agent-init guide flow`), then remove"
264481
+ );
264482
+ push(" # this marker.");
264483
+ }
264407
264484
  push(
264408
264485
  " # Choose repositories separately from ordinary files/folders alongside them."
264409
264486
  );
@@ -264637,6 +264714,9 @@ var init_seed = __esm({
264637
264714
  "# they require. Setup saves a fresh baseline, leaves the schedule off,"
264638
264715
  );
264639
264716
  push("# and does not run the script immediately.");
264717
+ push(
264718
+ "# Set timeoutSeconds to an integer from 1 through 480 (at most 8 minutes)."
264719
+ );
264640
264720
  push(
264641
264721
  "# Shape: { script, timeoutSeconds, watches: [{ repository, enabled, paths }] }"
264642
264722
  );
@@ -264949,7 +265029,7 @@ var init_status = __esm({
264949
265029
  nextAction = AGENT_INIT_CONNECT_AGENTS_ACTION;
264950
265030
  } else if (manifestText === null) {
264951
265031
  phase = "not_started";
264952
- nextAction = "Run `dvb agent-init start` to seed the manifest.";
265032
+ nextAction = "Choose the project folder (Phase 2 of `dvb agent-init guide flow`), then run `dvb agent-init start --project <path>` to seed the manifest.";
264953
265033
  } else if ((manifestTodoCount ?? 0) > 0) {
264954
265034
  phase = "composing";
264955
265035
  nextAction = `The manifest still has ${manifestTodoCount} TODO(agent) marker(s). Investigate each section (see \`dvb agent-init guide\`), then run \`dvb agent-init check\`.`;
@@ -297353,4 +297433,4 @@ react/cjs/react-jsx-runtime.development.js:
297353
297433
  *)
297354
297434
  */
297355
297435
 
297356
- //# debugId=e37678ce-fb44-5903-81f8-a25f3b4c4d9c
297436
+ //# debugId=b8e2b3d8-97c2-56a6-8a4f-a3626d87d781
package/dist/bin/dvbd.mjs CHANGED
@@ -64675,9 +64675,9 @@ var init_sentry = __esm({
64675
64675
  uncaughtExceptionMonitorInstalled = false;
64676
64676
  currentDevboxSentryUser = null;
64677
64677
  readBuildMetadata = () => {
64678
- const rawPackageVersion = "1.0.1072";
64679
- const rawGitSha = "38cca56e185da1e977f029154045c5ef6ebcef5a";
64680
- const rawSentryRelease = "boxes-dev-dvb@1.0.1072+38cca56e185da1e977f029154045c5ef6ebcef5a";
64678
+ const rawPackageVersion = "1.0.1074";
64679
+ const rawGitSha = "36b0586d995fab98581795868e96940b207d13b2";
64680
+ const rawSentryRelease = "boxes-dev-dvb@1.0.1074+36b0586d995fab98581795868e96940b207d13b2";
64681
64681
  const packageVersion = typeof rawPackageVersion === "string" ? rawPackageVersion : void 0;
64682
64682
  const gitSha = typeof rawGitSha === "string" ? rawGitSha : void 0;
64683
64683
  const sentryRelease = typeof rawSentryRelease === "string" ? rawSentryRelease : void 0;
@@ -66894,7 +66894,7 @@ var init_packageVersion = __esm({
66894
66894
  return path8.join(process.cwd(), "dvb");
66895
66895
  };
66896
66896
  readEmbeddedPackageVersion = () => {
66897
- const raw = "1.0.1072";
66897
+ const raw = "1.0.1074";
66898
66898
  return trimVersion(raw);
66899
66899
  };
66900
66900
  readNearestPackageMetadata = (basePath) => {
@@ -115,6 +115,11 @@ target `name` and `instanceId`. Use that `instanceId` directly with
115
115
  `--target`; do not refresh the full devbox inventory merely to resolve the
116
116
  name.
117
117
 
118
+ If creation fails, `--wait` exits with status 1 and includes the failure code
119
+ and message. JSON keeps `ok: true` for the successfully retrieved result;
120
+ check `data.operation.status` for the creation outcome. Reading that failed
121
+ operation with `devboxes status` exits successfully.
122
+
118
123
  For an asynchronous caller, omit `--wait`. Creation then returns immediately
119
124
  with a durable operation id, which can be polled with:
120
125
 
@@ -130,6 +135,10 @@ idempotency key only when the project is identical. A changed request under the
130
135
  same key fails. Do not generate a new key merely because a request timed out;
131
136
  first read the original operation.
132
137
 
138
+ After a confirmed terminal failure with `failure.retryable: true`, use a new
139
+ idempotency key for a new attempt. Reusing the previous key will return the
140
+ same failed operation.
141
+
133
142
  This workflow creates a devbox from the project's current reusable source
134
143
  without creating a draft, conversation, or agent thread. That source is an
135
144
  ordinary project's Template-box snapshot or a Team Template member project's
@@ -164,8 +164,9 @@ it is a complete agent selection rather than a partial merge:
164
164
  - Codex requires `backend`, `model`, and `reasoningEffort`. `serviceTier` is
165
165
  optional and defaults to `standard`; set it to `fast` to use Fast mode.
166
166
  - Claude Code requires `backend`, `model`, and `reasoningEffort`. Models
167
- without a reasoning control use `null`. Claude does not accept
168
- `serviceTier`.
167
+ without a reasoning control use `null`. `serviceTier` is optional and
168
+ defaults to `standard`; set it to `fast` to use fast mode on a model that
169
+ supports it.
169
170
 
170
171
  The values use the same model, reasoning, and speed controls as normal threads
171
172
  and automation defaults. Invalid fields, unknown fields, unsupported
@@ -29,10 +29,11 @@ The workbench has a few persistent areas:
29
29
  - The left pane lists threads, integrations, automations, Project URLs, an
30
30
  expandable Settings section, and the Template box when that project has one.
31
31
  Your account identity and a Help button stay pinned at the bottom.
32
- If you haven't signed in to the mobile app, a **Code on the go!** banner
33
- appears above your account. Scan its QR code to download the mobile app, or
34
- select **X** to dismiss it for this account on this computer. Signing in on
35
- mobile will also hide the banner.
32
+ Once you've sent messages in three threads (not counting project setup
33
+ threads), a **Code on the go!** banner appears above your account if you
34
+ haven't signed in to the mobile app. Scan its QR code to download the mobile
35
+ app, or select **X** to dismiss it for this account on this computer. Signing
36
+ in on mobile will also hide the banner.
36
37
  - The detail pane shows the selected thread, Project settings, an integration,
37
38
  an automation editor, or the Template box page when available.
38
39
  - The Terminal, Files, Git/Review, and Browser panes can be shown next to the
@@ -323,6 +324,21 @@ threads, new-devbox placeholders, and automation views all use the same detail
323
324
  area. A Codex or Claude thread can switch between GUI and the CLI from its
324
325
  header without becoming a different thread.
325
326
 
327
+ Use **Dictate** (the microphone beside the composer controls) to speak a
328
+ message to either agent. On first use, choose **Start dictation** and allow
329
+ microphone access. Your audio is sent to OpenAI to turn it into text. Words
330
+ appear live in your draft; click the microphone again to finish, then review,
331
+ edit, and Send normally. Dictation never sends a message automatically.
332
+ You can keep typing and adding attachments while speaking. The microphone is
333
+ highlighted while listening and shows a spinner while connecting or finishing.
334
+ Switching threads or hiding the window stops capture and keeps the text already
335
+ in your draft. An agent's Stop button remains separate.
336
+
337
+ Each recording stops after three minutes. Frequent use is limited to 120 starts
338
+ per account in an hour. If the connection fails, any words already received are
339
+ kept in your draft for review. Allow microphone access in your operating
340
+ system's settings if it was denied. Update the desktop app if Dictate is absent.
341
+
326
342
  Agent replies can include blue pills pointing to controls or settings. Hover
327
343
  one to see its location, then click to reveal and highlight the item. A
328
344
  reference can open a settings section or menu, but does not activate the
@@ -565,12 +565,21 @@ as Fix checks, Mark ready, Address feedback, Push to update PR, or Merge. Push
565
565
  to update PR uses the same Pull-first Push dialog. When **View PR #N** appears,
566
566
  it opens that pull request in your browser.
567
567
 
568
- Open the PR indicator in the conversation header and choose **Fix CI & apply
568
+ Hover over or click the PR indicator in the conversation header to see the
569
+ pull request's title, repository, branch, checks, review, and last refresh.
570
+ Threads using the same checkout share its branch and PR. If the checkout has
571
+ multiple pull requests, the panel lists each one with its own links.
572
+
573
+ Choose **Fix CI & apply
569
574
  feedback** to prepare the available CI fixes and review feedback together.
570
575
  Choose **Fix CI** or **Apply review feedback** to handle just one; an option
571
576
  is unavailable when there is no work of that kind. Review the prepared prompt
572
577
  and **Code review** attachment, then choose **Send** to start the agent.
573
- Choose **Open PR in Git/Review** in the same menu to inspect the pull request.
578
+ Choose **Open PR in Git/Review** in the same panel to inspect the pull request.
579
+
580
+ The collapsed Git/Review header shows checks and review separately. Choose
581
+ **Details** to expand it. **No checks reported** means GitHub hasn't reported
582
+ checks for that PR; it doesn't mean the checks passed.
574
583
 
575
584
  ## Automatic Pull Request Follow-ups
576
585
 
@@ -141,8 +141,9 @@ current thread state, primary transcript messages use the same phone-readable
141
141
  text size as the composer. The composer can also show:
142
142
 
143
143
  - a Codex or Claude agent selector before a draft starts;
144
- - a model control that opens the model, reasoning, and Codex speed
145
- (Standard, Fast, and Ultrafast when offered) choices in one menu;
144
+ - a model control that opens the model, reasoning, and speed choices in one
145
+ menu: Standard, Fast, and Ultrafast when a Codex model offers them, and
146
+ Standard or Fast (fast mode) for Claude models that support fast mode;
146
147
  - Claude Code model, reasoning, and permission mode choices in that same
147
148
  control when available;
148
149
  - context and rate-limit indicators;
@@ -151,6 +152,22 @@ text size as the composer. The composer can also show:
151
152
  - image and file attachments when available;
152
153
  - send, stop, and interrupt controls.
153
154
 
155
+ Tap **Dictate** (the microphone beside Send or the agent's Stop button) to
156
+ speak a message to Codex or Claude. On first use, tap **Start dictation** and
157
+ allow microphone access. Audio is sent to OpenAI to turn it into text. Words
158
+ appear live in your editable draft. Tap the microphone again to
159
+ finish, then review, edit, and Send normally. Dictation never sends
160
+ automatically and preserves attachments. The microphone is highlighted while
161
+ listening and shows a spinner while connecting or finishing. Leaving the thread or
162
+ putting the app in the background stops capture and keeps the draft's
163
+ text. The agent's Stop button still controls its work.
164
+
165
+ A recording stops after three minutes; frequent use is limited to 120 starts
166
+ per account in an hour. If the connection fails, words already received are
167
+ kept in the draft for review. If microphone access was denied, allow it for
168
+ boxes.dev in your phone's settings and try again. Dictate requires an updated
169
+ mobile app; an older installed app needs an app-store update.
170
+
154
171
  To find text within the open conversation, choose **… → Find in thread** in the
155
172
  thread header. **Find in thread** matches literal text, ignoring capitalization
156
173
  and Markdown formatting, in your messages and the agent's final replies. Use the
@@ -12,6 +12,12 @@ If a command described here is absent from help, the box has not received that
12
12
  capability yet. Do not install workstation `dvb`; wait for boxes.dev to update
13
13
  the box and try again.
14
14
 
15
+ `dvb-onbox` commands run through the local boxes.dev service on the box. When
16
+ boxes.dev updates the box, that service restarts, which usually takes a few
17
+ seconds, and commands wait up to 15 seconds for it to come back. If it's still
18
+ unavailable, the command fails with a message saying the local boxes.dev
19
+ service isn't available. Wait a minute and run the command again.
20
+
15
21
  Most permission-gated operations appear as focused agent skills only while
16
22
  their existing box or project setting is enabled. Turning one of those settings
17
23
  off removes its projected skill from both managed agent homes. The devbox
@@ -457,8 +463,9 @@ separate `--prompt-file`, not both. File paths can be relative or absolute.
457
463
  safeguards, and no outputs. `create` saves the automation paused.
458
464
  - `execution`: `backend` is `codex` or `claude`. Take `model` and
459
465
  `reasoningEffort` from `automations options`; `reasoningEffort` is `null`
460
- only for a model without reasoning control. Codex also takes `serviceTier`
461
- `standard` (the default) or `fast`.
466
+ only for a model without reasoning control. Both agents also take
467
+ `serviceTier` `standard` (the default) or `fast`; for Claude, `fast` is fast
468
+ mode and needs a model that supports it.
462
469
  - `trigger`: `{"kind": "manual"}`, or a schedule with `frequency` `hourly`,
463
470
  `daily`, `weekly`, or `monthly`; a whole-number `interval`; an IANA
464
471
  `timeZone`; a `startDate` (`YYYY-MM-DD`); `hour` and `minute`; `weekdays`
@@ -843,8 +850,9 @@ dvb-onbox threads options --agent claude --json
843
850
  ```
844
851
 
845
852
  The result includes exact model IDs, the default model, each model's available
846
- reasoning efforts, and its default effort. Codex models also list their
853
+ reasoning efforts, and its default effort. Each model also lists its
847
854
  available speeds: Standard, plus Fast and Ultrafast when the model offers them.
855
+ Claude offers Fast (fast mode) on models that support it and never Ultrafast.
848
856
  Treat it as the authority for the current box; do not guess from an older
849
857
  list. Discovery reads only the source box and does not create, wake, or
850
858
  inspect a destination devbox.
@@ -887,7 +895,7 @@ dvb-onbox threads start \
887
895
  ```
888
896
 
889
897
  To make an explicit selection, add a model returned by `threads options` and,
890
- when supported, one of that model's reasoning efforts and a Codex speed:
898
+ when supported, one of that model's reasoning efforts and speeds:
891
899
 
892
900
  ```bash
893
901
  dvb-onbox threads start \
@@ -904,12 +912,11 @@ dvb-onbox threads start \
904
912
 
905
913
  `--model` without `--reasoning-effort` uses that model's reported default
906
914
  effort. Omit `--reasoning-effort` for a model whose effort list is empty.
907
- `--speed` is Codex-only and accepts `fast` or `ultrafast` only for a model
908
- that reports that speed; an explicit model without speed uses Standard.
909
- Reasoning effort or speed without a model is invalid. When all execution
910
- controls are omitted, the child inherits the source box's saved defaults for
911
- the selected agent: the Codex model, reasoning, and speed, or the Claude model
912
- and reasoning.
915
+ `--speed` accepts `fast` or `ultrafast` only for a model that reports that
916
+ speed; an explicit model without speed uses Standard. Reasoning effort or speed
917
+ without a model is invalid. When all execution controls are omitted, the child
918
+ inherits the source box's saved model, reasoning, and speed defaults for the
919
+ selected agent.
913
920
  If that source box has no saved defaults for the agent, the managed catalog
914
921
  default is used.
915
922
 
@@ -1133,8 +1140,9 @@ names. Project filters apply to `user`, `shared`, and `team`; owner filters
1133
1140
  apply to `shared` and `team`. Each listing page examines up to 50 candidate
1134
1141
  records; filters and access checks can leave fewer results or an empty page.
1135
1142
  Pass `pagination.nextCursor` as `--cursor` when you need more results, and
1136
- check `pagination.complete` to know when the inventory ends. Scoped commands
1137
- do not accept `--all`. Project discovery can repeat a project across pages;
1143
+ check `pagination.complete` to know when the inventory ends. Add `--all` to a
1144
+ scoped thread listing to fetch every page in one command. It cannot be combined
1145
+ with `--cursor`. Project discovery can repeat a project across pages;
1138
1146
  deduplicate by project ID. Add `--include-archived` to thread listings for archived work.
1139
1147
 
1140
1148
  `read --thread` accepts a thread's `id` or `publicId` from a listing, a
@@ -1142,6 +1150,30 @@ boxes.dev thread link, or the thread's Codex thread ID or Claude session ID.
1142
1150
  Listings and reads report that agent ID as `nativeThreadId`. If an agent ID
1143
1151
  matches more than one thread you can read, use the thread's `id` instead.
1144
1152
 
1153
+ Listings, reads, and search results report the thread's last-reported execution
1154
+ `status`: `draft`, `starting`, `idle`, `running`, `needs_input`, `failed`, or
1155
+ `archived`. They also include `activeTurnId` and `activeTurnStatus` (both
1156
+ nullable), alongside `updatedAt`. Check the thread's `status` when deciding
1157
+ whether to nudge an idle agent; `transcript.status: ready` only means its
1158
+ conversation can be read. These stored states can lag live work or change
1159
+ before a message arrives, so they do not guarantee idle-only delivery. Reading
1160
+ status does not wake a devbox or require delegation permission.
1161
+
1162
+ For known threads, fetch their metadata together instead of scanning the whole
1163
+ project or reading their conversations:
1164
+
1165
+ ```bash
1166
+ dvb-onbox threads status --scope project --ids-file lanes.json --json
1167
+ ```
1168
+
1169
+ The file must contain a JSON array of 1 through 100 thread `id` values from
1170
+ a listing, such as `["<thread-id>", "<another-thread-id>"]`, and be at most
1171
+ 32 KiB. The same read scopes and permissions apply. Results follow the file's
1172
+ order, with `status: "available"` and the usual `thread` metadata, or
1173
+ `status: "unavailable"` with `thread: null` for an ID that cannot be read in
1174
+ that scope. Use `thread.status` for execution state. This lookup does not read
1175
+ conversations or wake devboxes, and its states may lag live work.
1176
+
1145
1177
  Listings, reads, and search results also report `devboxId` for your own
1146
1178
  threads on a devbox. It is null for teammates' threads and for threads on the
1147
1179
  Template box. With **Create devboxes and threads** enabled, pass it to
@@ -1206,8 +1238,42 @@ boxes need it.
1206
1238
  ```bash
1207
1239
  dvb-onbox threads message --to <thread-id> \
1208
1240
  --message "The endpoint now returns 422 for invalid widgets." --json
1241
+
1242
+ dvb-onbox threads message --to <thread-id> --message "Review the endpoint." \
1243
+ --idempotency-key keeper-2026-10-14-lane-1 --idle-only --json
1244
+
1245
+ dvb-onbox threads message --batch-file /tmp/messages.json --idle-only --json
1209
1246
  ```
1210
1247
 
1248
+ The batch file is a JSON array of 1–100 objects, up to 3 MiB:
1249
+
1250
+ ```json
1251
+ [
1252
+ {"to": "<thread-id>", "message": "Review the endpoint.", "idempotencyKey": "keeper-2026-10-14-lane-1"}
1253
+ ]
1254
+ ```
1255
+
1256
+ Each message needs a distinct key of 1–128 characters. The entire file is
1257
+ validated before sending. Four deliveries run concurrently, launching in file
1258
+ order, so put higher-priority threads first. Results retain that order even
1259
+ when deliveries finish in a different order. Each item succeeds or fails
1260
+ independently; a failed item does not cancel the others. JSON output includes
1261
+ `results`; the command exits unsuccessfully if any item fails.
1262
+
1263
+ `--idle-only` checks the recipient after startup, immediately before native
1264
+ delivery. It returns `delivered` for an accepted message or `skipped_busy`
1265
+ without adding input when the agent is busy or awaiting an answer. Skips count
1266
+ as successful outcomes. A status lookup can lag the runtime; use this flag to
1267
+ enforce idle admission. Sleeping recipients may wake before checking. This
1268
+ mode requires an updated recipient service; `unsupported_capability` means
1269
+ the recipient cannot enforce it and the message was not delivered.
1270
+
1271
+ Keys reuse the same receipt on the same source machine, including across CLI
1272
+ invocations. Retrying with the same key and message recovers that outcome
1273
+ without another delivery. Changing the recipient, message, or idle-only policy
1274
+ under that key fails with `idempotency_key_reused`. Use a new key for each new
1275
+ logical nudge or cadence pass, including when reconsidering a busy skip.
1276
+
1211
1277
  `--to` takes a boxes.dev thread ID, such as one named in a received message,
1212
1278
  one returned by `dvb-onbox threads start`, or one from
1213
1279
  `dvb-onbox threads list --scope project`. Use this command rather than a
@@ -1229,13 +1295,22 @@ messages; messages from the user and the agent's final answer stay visible.
1229
1295
 
1230
1296
  A successful result means the recipient accepted the message; it does not mean
1231
1297
  the agent has read, acted on, or answered it. Replies arrive later as messages
1232
- in the sender's thread. After `operation_indeterminate`, do not send the
1233
- message again, because it may already have arrived. A Claude agent in a
1298
+ in the sender's thread. After `operation_indeterminate`, recover with the same
1299
+ key if one was supplied; do not send with a new key, because it may already
1300
+ have arrived. Without a supplied key, do not send again. A Claude agent in a
1234
1301
  terminal that is busy or controlled by the user cannot receive messages
1235
1302
  (`thread_busy` or `thread_controlled_elsewhere`); an idle, unattended terminal
1236
1303
  session can. The recipient must be a started, unarchived thread in the same
1237
1304
  project; messaging teammates or other projects is not available.
1238
1305
 
1306
+ Sending can take a minute or more when the recipient's devbox has to wake or
1307
+ update first. Let the command finish: it waits up to about three minutes for
1308
+ the outcome before reporting `operation_indeterminate`, so a script that runs
1309
+ it should allow at least four minutes per delivery; batches may need longer
1310
+ because each group can include slow wakes. If a send is stopped early, recover
1311
+ with the same key or treat the unkeyed message as possibly delivered. Batch
1312
+ concurrency does not guarantee a five-minute cadence when machines must wake.
1313
+
1239
1314
  If sending reports `messaging_disabled`, request `message-agents`.
1240
1315
  `recipient_messaging_disabled` means the recipient's devbox has messaging off;
1241
1316
  only the user can change that, in that devbox's Agent permissions.
@@ -198,9 +198,14 @@ adapted for Linux without replacing the Template box's managed shell setup.
198
198
  Removing a file entry removes that explicit copy, but repository copying or
199
199
  the separate untracked-file selection may still include the file.
200
200
 
201
- For a folder containing several repositories, agent setup can select those
202
- repositories separately from ordinary files and folders alongside them.
203
- Automatic GitHub access remains the default for GitHub repositories. SSH setup
201
+ By default, agent setup creates a single-repository project. The coding agent
202
+ suggests including other repositories only when it finds evidence that the
203
+ user develops them together, such as one repository depending on another by
204
+ relative path, and it asks the user first. When the agent starts in a folder
205
+ of unrelated repositories, it asks which one to set up. For a folder
206
+ containing several repositories, agent setup can select those repositories
207
+ separately from ordinary files and folders alongside them. Automatic GitHub
208
+ access remains the default for GitHub repositories. SSH setup
204
209
  uses one key for the selected repositories that need it, including child
205
210
  repositories and repositories on other hosts; the key needs access on each
206
211
  host. Automatic GitHub access and SSH for another host can be used together.
@@ -360,12 +365,17 @@ GitHub and Slack. Codex and Claude Code use their normal browser sign-in in
360
365
  the same tab, which returns to the page through the CLI on the user's
361
366
  computer; neither agent needs to be installed there. **Use an API key** saves
362
367
  a key instead. **Done** becomes available once a coding agent is connected,
363
- and the command waits until the user chooses it. Over SSH, the command prints
368
+ and the command waits until the user chooses it. If an agent is already
369
+ connected to boxes.dev, no new sign-in is needed; choose **Done** to continue
370
+ setup with that account. Over SSH, the command prints
364
371
  a link for the user to open and exits: on that page, Codex shows a one-time
365
372
  code to enter on ChatGPT, Claude Code asks for the code Claude shows, and the
366
373
  user tells the agent when they're done. **Use a code instead** switches a
367
374
  local page to the same code flows for setups where browser sign-in can't
368
- reach the CLI, such as a container. `dvb agent-init check` requires at least
375
+ reach the CLI, such as a container. Choosing **Done** there also finishes the
376
+ waiting command. With an older CLI, tell the agent you clicked Done if it
377
+ keeps waiting; it can stop the command and check setup status.
378
+ `dvb agent-init check` requires at least
369
379
  one connected coding agent.
370
380
 
371
381
  Once the setup thread starts on the Template box, `dvb agent-init run` tells
@@ -404,13 +414,14 @@ user selects the key to switch to it.
404
414
  **Log in with Claude** saves a claude.ai login, the same kind of sign-in
405
415
  Claude Code gets from `/login` on a computer. Claude on Template boxes and
406
416
  devboxes then has the account's claude.ai features: skills and plugins synced
407
- from claude.ai and the connectors the user added there, such as Gmail.
417
+ from claude.ai, the connectors the user added there, such as Gmail, and fast
418
+ mode.
408
419
  boxes.dev keeps the sign-in current, but Claude requires a new sign-in about
409
420
  every four weeks; Claude sets the exact date at each sign-in. Three days
410
421
  before that deadline, the account's row shows **Sign in again by** a date and
411
422
  the mobile app shows a reminder banner; after it passes, Claude reports that
412
423
  the sign-in expired until the user signs in again. Accounts connected before claude.ai logins were available show
413
- **Setup token**. They keep working without claude.ai features, and the row's
424
+ **Setup token**. They keep working without claude.ai features or fast mode, and the row's
414
425
  **Switch to claude.ai login** action signs in again and converts that saved
415
426
  account to a claude.ai login. An older account saved without a Claude account
416
427
  ID gets a new row for the login instead, which becomes active; the user can
@@ -440,7 +451,9 @@ subscription radio chooses the **Default** account while automatic selection
440
451
  is enabled; **Default** appears only while an automatic option is on, and
441
452
  **Active** marks the account currently in use. boxes.dev prefers
442
453
  the default whenever its usage is available, switches to another subscription
443
- when it is exhausted, and returns after quota resets. A saved API key also
454
+ when it is exhausted, and returns after quota resets; for Codex, only once a
455
+ fresh reading of the default confirms it (an account that cannot be read is
456
+ assumed reset). A saved API key also
444
457
  exposes **Automatically switch to API when all subscriptions are exhausted**,
445
458
  which requires a separate opt-in and incurs the API provider's charges. Unknown
446
459
  quota and authentication failures alone do not authorize API switching.
@@ -449,9 +462,16 @@ on either automatic option while an API key is selected moves the agent to a
449
462
  subscription unless every subscription is exhausted and API switching is on.
450
463
  Codex and Claude settings are independent, global, and affect conversations started from either desktop or mobile. Mobile can
451
464
  still select accounts manually; these automatic controls are configured on
452
- desktop. Account changes retain conversation history; an agent already working
453
- may finish its current turn first. They do not automatically replay a turn that
454
- stopped at a usage limit; the user can continue that conversation after switching.
465
+ desktop. Account changes retain conversation history. A Codex turn that is
466
+ running when its account changes stops, and boxes.dev continues it on the new
467
+ account with an empty turn about five seconds later; its failure card says so.
468
+ If the new account is out of usage too, a turn that hits its limit waits up to
469
+ a minute and continues on the next account rotation selects. A Codex turn that
470
+ stopped at a usage limit before the switch still needs a follow-up. When Claude hits a usage limit and automatic selection finds another
471
+ account, boxes.dev resumes the same conversation and continues the unfinished
472
+ request before queued follow-ups. Stop cancels this recovery. With no available
473
+ account, or with background work or pending steering that cannot be safely
474
+ resumed, the turn stays stopped.
455
475
 
456
476
  Configure agents manages the Codex and Claude accounts that run coding-agent
457
477
  threads. Remote MCP connections are managed separately from **Integrations →
@@ -41,9 +41,9 @@ devbox. Archived conversations need to be unarchived in boxes.dev first;
41
41
  conversations in storage need to be restored. A history-only conversation
42
42
  cannot continue. These recovery actions are not performed automatically.
43
43
 
44
- A new task uses the agent's saved model and reasoning effort for the project,
45
- and its saved speed for Codex, unless the user asks for a specific model,
46
- reasoning effort or Codex speed. The assistant can choose from the models the
44
+ A new task uses the agent's saved model, reasoning effort and speed for the
45
+ project, unless the user asks for a specific model, reasoning effort or speed.
46
+ Claude's fast speed is fast mode. The assistant can choose from the models the
47
47
  boxes.dev model picker lists for the account, and the start receipt reports
48
48
  what the agent was launched with.
49
49
 
@@ -756,6 +756,19 @@ enabled. On desktop, `/model` opens the model picker, and `/model <model>`
756
756
  switches directly to a known model. Older supported model generations appear
757
757
  under **More models** in the picker.
758
758
 
759
+ Fast mode runs Claude faster at a higher usage rate on models that support it,
760
+ such as Opus 5.5. On desktop, the model menu has an **Enable fast mode** switch
761
+ and shows a lightning icon beside the model name while fast mode is on; `/fast`
762
+ toggles it, and `/fast on` or `/fast off` sets it. Mobile lists it as **Fast**
763
+ under **Speed**. The CLI agent view, automations, delegated threads, and remote
764
+ assistants accept it as the `fast` speed. Fast mode needs extra usage turned on
765
+ in the Claude account and a Claude account connected as a claude.ai login; an
766
+ account labeled **Setup token** needs **Switch to claude.ai login** first. When
767
+ Claude reports fast mode as unavailable, for example because extra usage is off,
768
+ the account uses a setup token, the organization has turned it off, or a fast
769
+ mode rate limit is in effect, the menu shows the reason and Claude runs at
770
+ standard speed.
771
+
759
772
  An idle Claude chat can continue in a separate chat. Use exact `/fork` to
760
773
  continue from the current tip on the current Template box or devbox. Or use the
761
774
  Split icon beneath a completed desktop response to keep history only through
@@ -770,7 +783,7 @@ for managed Claude chats that support chat forking. Desktop remembers confirmed
770
783
  support for each Template box and devbox and keeps the controls available while
771
784
  reconnecting; choosing a fork action will connect to the source first.
772
785
 
773
- Model and reasoning selections belong to the Claude thread. A new
786
+ Model, reasoning, and fast mode selections belong to the Claude thread. A new
774
787
  selection remains associated with that thread while it is applied, and the
775
788
  confirmed settings follow the thread when it is reopened or resumed on desktop
776
789
  or mobile. Changing one thread does not change another existing thread; the
@@ -224,8 +224,12 @@ Codex failure cards distinguish several failure layers so the recovery step
224
224
  matches what happened:
225
225
 
226
226
  - Selected-model capacity and temporary provider throttling come from the model
227
- provider, not the selected machine. Continue after a short wait, or choose
228
- another model if capacity remains limited.
227
+ provider, not the selected machine. When the selected model is at capacity,
228
+ boxes.dev continues the thread automatically after a short wait, with longer
229
+ waits if capacity stays limited, and sends no failure notification while it
230
+ does. If the card asks you to reply, boxes.dev has stopped trying: continue
231
+ after a short wait, or choose another model. For temporary throttling,
232
+ continue after a short wait.
229
233
  - Codex credits and authentication are account problems. Add or wait for
230
234
  credits, or reconnect Codex before continuing.
231
235
  - A box networking card means the selected machine could not reach the model
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@boxes-dev/dvb-runtime",
3
- "version": "1.0.1072",
3
+ "version": "1.0.1074",
4
4
  "description": "Runtime for the boxes.dev CLI. Install @boxes-dev/dvb to use the CLI.",
5
5
  "homepage": "https://boxes.dev",
6
6
  "type": "module",
@@ -77,5 +77,5 @@
77
77
  "e2b": "^2.18.0",
78
78
  "ink-testing-library": "^4.0.0"
79
79
  },
80
- "devboxReleasedAt": "2026-10-14T03:26:02Z"
80
+ "devboxReleasedAt": "2026-10-14T15:37:29Z"
81
81
  }