@vellumai/assistant 0.4.56 → 0.5.0

package/src/tools/watcher/list.ts

@@ -75,7 +75,7 @@ export async function executeWatcherList(
       ? new Date(w.lastPollAt).toLocaleString()
       : "never";
     lines.push(
-      `  - [${status}] ${w.name} (${w.providerId}) — last: ${lastPoll}`,
+      `  - [${status}] ${w.name} (${w.providerId}) - last: ${lastPoll}`,
     );
   }
 

package/src/util/logger.ts

@@ -12,7 +12,10 @@ import pino from "pino";
 import type { PrettyOptions } from "pino-pretty";
 import pinoPretty from "pino-pretty";
 
-import { getDebugStdoutLogs } from "../config/env-registry.js";
+import {
+  getDebugStdoutLogs,
+  getIsContainerized,
+} from "../config/env-registry.js";
 import { logSerializers } from "./log-redact.js";
 import { getLogPath } from "./platform.js";
 
@@ -109,7 +112,9 @@ function buildRotatingLogger(config: LogFileConfig): pino.Logger {
   // When stdout is not a TTY (e.g. desktop app redirects to a hatch log file),
   // write to the rotating file only — the hatch log already captured early
   // startup output and echoing pino output there is unnecessary duplication.
-  if (!process.stdout.isTTY) {
+  // Exception: in containers, always write to stdout so `docker logs` can
+  // capture daemon output for debugging.
+  if (!process.stdout.isTTY && !getIsContainerized()) {
     return pino(
       { name: "assistant", level: "info", serializers: logSerializers },
       fileStream,

package/src/util/pricing.ts

@@ -24,6 +24,9 @@ const PROVIDER_PRICING: Record<string, Record<string, ModelPricing>> = {
     "claude-haiku-4": { inputPer1M: 0.8, outputPer1M: 4 },
   },
   openai: {
+    "gpt-5.4": { inputPer1M: 2.5, outputPer1M: 15 },
+    "gpt-5.4-nano": { inputPer1M: 0.2, outputPer1M: 1.25 },
+    "gpt-5.2": { inputPer1M: 1.75, outputPer1M: 14 },
     "gpt-4o": { inputPer1M: 2.5, outputPer1M: 10 },
     "gpt-4o-mini": { inputPer1M: 0.15, outputPer1M: 0.6 },
     "gpt-4.1": { inputPer1M: 2.0, outputPer1M: 8.0 },
@@ -35,6 +38,7 @@ const PROVIDER_PRICING: Record<string, Record<string, ModelPricing>> = {
     "o4-mini": { inputPer1M: 1.1, outputPer1M: 4.4 },
   },
   gemini: {
+    "gemini-3-flash": { inputPer1M: 0.5, outputPer1M: 3 },
     "gemini-2.5-pro": { inputPer1M: 1.25, outputPer1M: 10 },
     "gemini-2.5-flash": { inputPer1M: 0.15, outputPer1M: 0.6 },
     "gemini-2.0-flash": { inputPer1M: 0.1, outputPer1M: 0.4 },

package/src/util/retry.ts

@@ -74,9 +74,19 @@ export function isRetryableStatus(status: number): boolean {
   return status === 429 || status >= 500;
 }
 
+/**
+ * Message patterns that indicate a retryable network/transport error even
+ * when no errno code is present (e.g. Bun's native fetch socket errors).
+ */
+const RETRYABLE_NETWORK_MESSAGE_PATTERNS = [
+  /socket.*closed unexpectedly/i,
+  /socket hang up/i,
+];
+
 /**
  * Whether an error is a retryable network error (ECONNRESET, ECONNREFUSED, etc.).
- * Checks both the error itself and one level of `cause` chain.
+ * Checks errno codes on the error and one level of `cause` chain, then falls
+ * back to message-based detection for runtime-specific errors (e.g. Bun fetch).
  */
 export function isRetryableNetworkError(error: unknown): boolean {
   if (!(error instanceof Error)) return false;
@@ -96,6 +106,24 @@ export function isRetryableNetworkError(error: unknown): boolean {
     if (causeCode && retryableCodes.has(causeCode)) return true;
   }
 
+  // Fall back to message-based detection for errors without errno codes
+  // (e.g. Bun's "The socket connection was closed unexpectedly")
+  if (
+    RETRYABLE_NETWORK_MESSAGE_PATTERNS.some((p) => p.test(error.message))
+  ) {
+    return true;
+  }
+
+  // Also check the cause's message (ProviderError wraps the original message
+  // but the cause retains the raw transport-level text)
+  const cause = error.cause;
+  if (
+    cause instanceof Error &&
+    RETRYABLE_NETWORK_MESSAGE_PATTERNS.some((p) => p.test(cause.message))
+  ) {
+    return true;
+  }
+
   return false;
 }
 

package/src/workspace/migrations/007-web-search-provider-rename.ts

@@ -0,0 +1,37 @@
+import { existsSync, readFileSync, writeFileSync } from "node:fs";
+import { join } from "node:path";
+
+import type { WorkspaceMigration } from "./types.js";
+
+export const webSearchProviderRenameMigration: WorkspaceMigration = {
+  id: "007-web-search-provider-rename",
+  description:
+    'Rename web-search provider from "anthropic-native" to "inference-provider-native"',
+  run(workspaceDir: string): void {
+    const configPath = join(workspaceDir, "config.json");
+    if (!existsSync(configPath)) return;
+
+    let config: Record<string, unknown>;
+    try {
+      const raw = JSON.parse(readFileSync(configPath, "utf-8"));
+      if (!raw || typeof raw !== "object" || Array.isArray(raw)) return;
+      config = raw as Record<string, unknown>;
+    } catch {
+      return;
+    }
+
+    const services = config.services;
+    if (!services || typeof services !== "object" || Array.isArray(services))
+      return;
+
+    const webSearch = (services as Record<string, unknown>)["web-search"];
+    if (!webSearch || typeof webSearch !== "object" || Array.isArray(webSearch))
+      return;
+
+    const ws = webSearch as Record<string, unknown>;
+    if (ws.provider !== "anthropic-native") return;
+
+    ws.provider = "inference-provider-native";
+    writeFileSync(configPath, JSON.stringify(config, null, 2) + "\n");
+  },
+};

package/src/workspace/migrations/registry.ts

@@ -4,6 +4,7 @@ import { seedDeviceIdMigration } from "./003-seed-device-id.js";
 import { extractCollectUsageDataMigration } from "./004-extract-collect-usage-data.js";
 import { addSendDiagnosticsMigration } from "./005-add-send-diagnostics.js";
 import { servicesConfigMigration } from "./006-services-config.js";
+import { webSearchProviderRenameMigration } from "./007-web-search-provider-rename.js";
 import type { WorkspaceMigration } from "./types.js";
 
 /**
@@ -17,4 +18,5 @@ export const WORKSPACE_MIGRATIONS: WorkspaceMigration[] = [
   extractCollectUsageDataMigration,
   addSendDiagnosticsMigration,
   servicesConfigMigration,
+  webSearchProviderRenameMigration,
 ];

package/src/__tests__/cli-help-reference-sync.test.ts

@@ -1,26 +0,0 @@
-import { describe, expect, test } from "bun:test";
-
-import { buildCliProgram } from "../cli/program.js";
-import { CLI_HELP_REFERENCE } from "../cli/reference.js";
-
-/**
- * Guard test: CLI_HELP_REFERENCE must stay in sync with the actual CLI help
- * output produced by buildCliProgram().helpInformation().
- *
- * The snapshot in reference.ts is embedded in the system prompt (via
- * system-prompt.ts) so the assistant knows which CLI commands are available.
- * If the actual CLI program drifts from the snapshot, the system prompt will
- * contain stale information.
- *
- * When this test fails, update CLI_HELP_REFERENCE in
- * assistant/src/cli/reference.ts to match the current output of
- * buildCliProgram().helpInformation().
- */
-describe("CLI_HELP_REFERENCE sync", () => {
-  test("CLI_HELP_REFERENCE matches buildCliProgram().helpInformation()", () => {
-    const program = buildCliProgram();
-    const actual = program.helpInformation();
-
-    expect(actual).toBe(CLI_HELP_REFERENCE);
-  });
-});

package/src/__tests__/onboarding-starter-tasks.test.ts

@@ -1,190 +0,0 @@
-import { existsSync, mkdirSync, rmSync, writeFileSync } from "node:fs";
-import { tmpdir } from "node:os";
-import { join } from "node:path";
-import { afterEach, beforeEach, describe, expect, test } from "bun:test";
-
-const TEST_DIR = join(
-  tmpdir(),
-  `vellum-starter-tasks-test-${crypto.randomUUID()}`,
-);
-
-import { mock } from "bun:test";
-
-// eslint-disable-next-line @typescript-eslint/no-require-imports
-const realPlatform = require("../util/platform.js");
-mock.module("../util/platform.js", () => ({
-  ...realPlatform,
-  getRootDir: () => TEST_DIR,
-  getDataDir: () => TEST_DIR,
-  getWorkspaceDir: () => TEST_DIR,
-  getWorkspaceConfigPath: () => join(TEST_DIR, "config.json"),
-  getWorkspaceSkillsDir: () => join(TEST_DIR, "skills"),
-  getWorkspaceHooksDir: () => join(TEST_DIR, "hooks"),
-  getWorkspacePromptPath: (file: string) => join(TEST_DIR, file),
-  ensureDataDir: () => {},
-  getPidPath: () => join(TEST_DIR, "vellum.pid"),
-  getDbPath: () => join(TEST_DIR, "data", "assistant.db"),
-  getLogPath: () => join(TEST_DIR, "logs", "vellum.log"),
-  getHistoryPath: () => join(TEST_DIR, "history"),
-  getHooksDir: () => join(TEST_DIR, "hooks"),
-
-  getSandboxRootDir: () => join(TEST_DIR, "sandbox"),
-  getSandboxWorkingDir: () => TEST_DIR,
-  getInterfacesDir: () => join(TEST_DIR, "interfaces"),
-  isMacOS: () => process.platform === "darwin",
-  isLinux: () => process.platform === "linux",
-  isWindows: () => process.platform === "win32",
-  getPlatformName: () => process.platform,
-  getClipboardCommand: () => null,
-  readSessionToken: () => null,
-  readLockfile: () => null,
-  writeLockfile: () => {},
-}));
-
-const noopLogger = new Proxy({} as Record<string, unknown>, {
-  get: (_target, prop) => (prop === "child" ? () => noopLogger : () => {}),
-});
-
-// eslint-disable-next-line @typescript-eslint/no-require-imports
-const realLogger = require("../util/logger.js");
-mock.module("../util/logger.js", () => ({
-  ...realLogger,
-  getLogger: () => noopLogger,
-  getCliLogger: () => noopLogger,
-  truncateForLog: (v: string) => v,
-  initLogger: () => {},
-  pruneOldLogFiles: () => 0,
-}));
-
-const { buildStarterTaskPlaybookSection, buildSystemPrompt } =
-  await import("../prompts/system-prompt.js");
-
-describe("buildStarterTaskPlaybookSection", () => {
-  test("returns a string with the section heading", () => {
-    const section = buildStarterTaskPlaybookSection();
-    expect(section).toContain("## Starter Task Playbooks");
-  });
-
-  test("documents all three kickoff intents", () => {
-    const section = buildStarterTaskPlaybookSection();
-    expect(section).toContain("[STARTER_TASK:make_it_yours]");
-    expect(section).toContain("[STARTER_TASK:research_topic]");
-    expect(section).toContain("[STARTER_TASK:research_to_ui]");
-  });
-
-  test("includes the make_it_yours playbook", () => {
-    const section = buildStarterTaskPlaybookSection();
-    expect(section).toContain("### Playbook: make_it_yours");
-    expect(section).toContain("accent color");
-    expect(section).toContain("Color Preference");
-    expect(section).toContain("user_selected");
-  });
-
-  test("make_it_yours uses app_file_edit instead of invalid ui_show config_update", () => {
-    const section = buildStarterTaskPlaybookSection();
-    expect(section).toContain("app_file_edit");
-    expect(section).not.toContain("config_update");
-    expect(section).not.toContain('surface_type: "config_update"');
-  });
-
-  test("includes the research_topic playbook", () => {
-    const section = buildStarterTaskPlaybookSection();
-    expect(section).toContain("### Playbook: research_topic");
-    expect(section).toContain("web search");
-  });
-
-  test("includes the research_to_ui playbook", () => {
-    const section = buildStarterTaskPlaybookSection();
-    expect(section).toContain("### Playbook: research_to_ui");
-    expect(section).toContain("app_create");
-  });
-
-  test("includes general rules section", () => {
-    const section = buildStarterTaskPlaybookSection();
-    expect(section).toContain("### General rules for all starter tasks");
-  });
-
-  test("enforces trust gating in general rules", () => {
-    const section = buildStarterTaskPlaybookSection();
-    expect(section).toContain("trust gating");
-    expect(section).toContain("do NOT ask for elevated permissions");
-  });
-
-  test("references USER.md onboarding tasks for status tracking", () => {
-    const section = buildStarterTaskPlaybookSection();
-    expect(section).toContain("## Onboarding Tasks");
-    expect(section).toContain("in_progress");
-    expect(section).toContain("done");
-    expect(section).toContain("deferred_to_dashboard");
-  });
-
-  test("make_it_yours playbook handles locale confirmation", () => {
-    const section = buildStarterTaskPlaybookSection();
-    expect(section).toContain("locale");
-    expect(section).toContain("confidence: low");
-  });
-
-  test("make_it_yours playbook includes confirmation step", () => {
-    const section = buildStarterTaskPlaybookSection();
-    expect(section).toContain("Confirm the selection");
-    expect(section).toContain("Sound good?");
-  });
-
-  test("research_to_ui playbook references dynamic UI quality standards", () => {
-    const section = buildStarterTaskPlaybookSection();
-    expect(section).toContain("Dynamic UI quality standards");
-    expect(section).toContain("anti-AI-slop");
-  });
-});
-
-describe("starter task playbook integration with buildSystemPrompt", () => {
-  beforeEach(() => {
-    mkdirSync(TEST_DIR, { recursive: true });
-  });
-
-  afterEach(() => {
-    if (existsSync(TEST_DIR)) {
-      rmSync(TEST_DIR, { recursive: true, force: true });
-    }
-  });
-
-  test("buildSystemPrompt includes the starter task playbook section when BOOTSTRAP.md exists", () => {
-    writeFileSync(join(TEST_DIR, "IDENTITY.md"), "I am Vellum.");
-    writeFileSync(join(TEST_DIR, "BOOTSTRAP.md"), "# First run");
-    const result = buildSystemPrompt();
-    expect(result).toContain("## Starter Task Playbooks");
-  });
-
-  test("buildSystemPrompt omits starter task playbooks after onboarding is complete", () => {
-    writeFileSync(join(TEST_DIR, "IDENTITY.md"), "I am Vellum.");
-    // No BOOTSTRAP.md → onboarding complete
-    const result = buildSystemPrompt();
-    expect(result).not.toContain("## Starter Task Playbooks");
-  });
-
-  test("starter task playbook and channel awareness both present during onboarding", () => {
-    writeFileSync(join(TEST_DIR, "IDENTITY.md"), "I am Vellum.");
-    writeFileSync(join(TEST_DIR, "BOOTSTRAP.md"), "# First run");
-    const result = buildSystemPrompt();
-    const starterIdx = result.indexOf("## Starter Task Playbooks");
-    const channelIdx = result.indexOf("## Channel Awareness & Trust Gating");
-    expect(starterIdx).toBeGreaterThan(-1);
-    expect(channelIdx).toBeGreaterThan(-1);
-  });
-
-  test("all three kickoff intents present in full system prompt during onboarding", () => {
-    writeFileSync(join(TEST_DIR, "IDENTITY.md"), "I am Vellum.");
-    writeFileSync(join(TEST_DIR, "BOOTSTRAP.md"), "# First run");
-    const result = buildSystemPrompt();
-    expect(result).toContain("[STARTER_TASK:make_it_yours]");
-    expect(result).toContain("[STARTER_TASK:research_topic]");
-    expect(result).toContain("[STARTER_TASK:research_to_ui]");
-  });
-
-  test("system prompt does not contain invalid config_update surface type (bare)", () => {
-    writeFileSync(join(TEST_DIR, "IDENTITY.md"), "I am Vellum.");
-    const result = buildSystemPrompt();
-    // voice_config_update is a valid tool name; only bare 'config_update' surface type is invalid
-    expect(result).not.toContain('surface_type: "config_update"');
-  });
-});

package/src/cli/reference.ts

@@ -1,38 +0,0 @@
-// Keep this snapshot in sync with buildCliProgram().helpInformation() for the
-// top-level assistant CLI. It deliberately avoids importing the CLI command
-// graph so prompt assembly stays side-effect-free.
-export const CLI_HELP_REFERENCE = `Usage: assistant [options] [command]
-
-Local AI assistant
-
-Options:
-  -V, --version                            output the version number
-  -h, --help                               display help for command
-
-Commands:
-  bash [options] <command>                 Execute a shell command through the assistant process for debugging
-  conversations                            Manage conversations
-  config                                   Manage configuration
-  keys                                     Manage API keys in secure storage
-  credentials [options]                    Manage credentials in the encrypted vault (API keys, tokens, passwords)
-  credential-execution [options]           Inspect and manage Credential Execution Service (CES) grants and audit records
-  trust                                    Manage trust rules
-  memory                                   Manage long-term memory indexing/retrieval
-  audit [options]                          Show recent tool invocations
-  avatar                                   Manage the assistant's avatar
-  doctor                                   Run diagnostic checks
-  hooks                                    Manage hooks
-  mcp                                      Manage MCP (Model Context Protocol) servers
-  contacts [options]                       Manage and query the contact graph
-  channel-verification-sessions [options]  Manage channel verification sessions
-  autonomy [options]                       View and configure autonomy tiers
-  completions <shell>                      Generate shell completion script (e.g. assistant completions bash >> ~/.bashrc)
-  notifications [options]                  Send and inspect notifications through the unified notification router
-  platform [options]                       Manage platform integration for containerized deployments
-  oauth [options]                          Manage OAuth providers, apps, connections, and tokens
-  skills                                   Browse and install skills from the Vellum catalog
-  browser                                  Browser automation, extension relay, and Chrome CDP management
-  usage                                    Query LLM token usage and cost data
-  shotgun                                  Start and monitor screen-watch (shotgun) sessions via IPC
-  sequence [options]                       Manage email sequences
-`;