Merge pull request #198 from modelstudioai/codex/release-1.23.0

chore(release): prepare 1.23.0

CHANGELOG.md

@@ -6,6 +6,44 @@ The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and
 
 [中文版](CHANGELOG.zh.md) · [README](README.md) · [Contributing](CONTRIBUTING.md)
 
+## [1.23.0] - 2026-09-10
+
+### Added
+
+- **Profile-level watermark control** — configure `watermark` with `bl config set --key watermark --value true|false` to control the default watermark behavior for image generation and editing, video generation and editing, and reference-to-video commands.
+- **ASR accuracy controls** — `bl speech recognize` now supports instant hot words with `--vocabulary`, contextual word enhancement with `--context`, and reusable pre-built vocabularies with `--vocabulary-id` for supported ASR models.
+- **Speech vocabulary management** — added `bl speech vocabulary create|list|get|update|delete` to manage reusable pre-built hot-word vocabularies.
+
+## [1.22.0] - 2026-09-08
+
+### Changed
+
+- **Project initialization** — `managed-agent project init` now creates `./managed-agent` by default. Use `--project .` to initialize in place. **(BREAKING)**
+- **Build confirmation** — `managed-agent project build` no longer requires confirmation and rejects `--yes`. Use `--dry-run` for a read-only preview; Publish still requires confirmation. **(BREAKING)**
+- **Managed Agent SDK** — upgrade to `0.7.1`. Build automatically associates active Agent-local resources while preserving explicit bindings, Skill versions, and File mount paths. Ambiguous Environment or Vault selections are rejected before writing.
+
+### Fixed
+
+- **Project diagnostics** — provide actionable project-root guidance and surface the underlying Build validation error.
+- **YAML initialization paths** — show the absolute YAML path in creation messages and existing-file errors.
+
+### Internal
+
+- Expand project initialization and Build regression coverage, and remove the obsolete Build confirmation flag from the local lifecycle E2E test.
+
+## [1.21.0] - 2026-09-07
+
+### Added
+
+- **Managed Agent directory projects** — initialize, validate, build, and publish local projects with Agent-scoped resource configuration and resource examples.
+- **Local project versions** — enable or disable snapshot versioning, list and preview history, and restore project files without Git.
+- **Project Workbench** — edit resources, review changes, publish updates, and manage local versions in a browser.
+
+### Changed
+
+- **Playground version resolution** — check npm before launch, reuse matching local versions, and fetch the latest version when needed; retain explicit version and binary overrides.
+- **Managed Agent SDK** — upgrade to `0.7.0` and consume project workspace and version services through SDK subpath exports.
+
 ## [1.20.0] - 2026-09-03
 
 > Managed Agents now combines YAML-first infrastructure management with direct Bailian AgentStudio resource and runtime operations.

CHANGELOG.zh.md

@@ -6,6 +6,44 @@
 
 [English](CHANGELOG.md) · [README](README.zh.md) · [参与贡献](CONTRIBUTING.zh.md)
 
+## [1.23.0] - 2026-09-10
+
+### 新增
+
+- **Profile 级水印控制** —— 可通过 `bl config set --key watermark --value true|false` 设置图片生成与编辑、视频生成与编辑以及参考生视频命令的默认水印行为。
+- **ASR 准确率增强** —— `bl speech recognize` 现支持通过 `--vocabulary` 传入即时热词、通过 `--context` 增强上下文词表，以及通过 `--vocabulary-id` 使用适用于对应 ASR 模型的预编译热词表。
+- **语音热词表管理** —— 新增 `bl speech vocabulary create|list|get|update|delete`，用于管理可复用的预编译热词表。
+
+## [1.22.0] - 2026-09-08
+
+### 变更
+
+- **项目初始化** —— `managed-agent project init` 默认创建 `./managed-agent` 子目录；如需原地初始化，请使用 `--project .`。**(BREAKING)**
+- **Build 确认机制** —— `managed-agent project build` 无需确认，并且不再接受 `--yes`。使用 `--dry-run` 可只读预览；Publish 仍需显式确认。**(BREAKING)**
+- **Managed Agent SDK** —— 升级至 `0.7.1`。Build 自动关联 Agent 目录下已启用的资源，保留显式引用、Skill 版本和 File 挂载路径；Environment 或 Vault 选择存在歧义时，在写入前报错。
+
+### 修复
+
+- **项目诊断** —— 提供可操作的项目根目录提示，并展示 Build 校验失败的具体原因。
+- **YAML 初始化路径** —— 创建成功及文件已存在的错误信息均展示 YAML 绝对路径。
+
+### 内部
+
+- 补充项目初始化和 Build 回归覆盖，移除本地闭环 E2E 测试中过时的 Build 确认参数。
+
+## [1.21.0] - 2026-09-07
+
+### 新增
+
+- **Managed Agent 目录项目** —— 支持初始化、校验、构建和发布本地项目，按 Agent 组织资源配置，并提供资源示例。
+- **本地项目版本管理** —— 无需 Git 即可启停快照版本管理、查看和预览历史，以及恢复项目文件。
+- **项目 Workbench** —— 在浏览器中编辑资源、审阅变更、发布更新和管理本地版本。
+
+### 变更
+
+- **Playground 版本选择** —— 启动前查询 npm，复用版本一致的本地安装，否则按需下载最新版；保留显式版本和启动文件覆盖。
+- **Managed Agent SDK** —— 升级至 `0.7.0`，通过 SDK 子路径复用项目目录和版本服务。
+
 ## [1.20.0] - 2026-09-03
 
 > Managed Agent 现在同时提供 YAML-first 基础设施管理与百炼 AgentStudio 资源、运行时 API 操作。

README.md

@@ -115,6 +115,7 @@ Once installed, just describe your task to your AI Agent — no need to assemble
 | ------------------------ | --------------------------------------------------------------------------------- |
 | Managed Agent            | "Create a Managed Agent that can generate short-film storyboards and videos."     |
 | Image & video generation | "Generate an image of a cat in a spacesuit on Mars, then turn it into a video."   |
+| Speech recognition       | "Transcribe this audio; if proper nouns are wrong, add hot words and try again."  |
 | Usage & quota            | "Show my recent model usage, free-tier quota, and rate limits."                   |
 | Model selection          | "Recommend a model for image understanding and customer support."                 |
 | About Bailian CLI        | "Tell me what Bailian CLI can do for me, and suggest how to use it for my needs." |

README.zh.md

@@ -114,6 +114,7 @@ irm https://bailian.aliyun.com/cli/install.ps1 | iex
 | ---------------- | ----------------------------------------------------------------------- |
 | Managed Agent    | “帮我创建一个能够生成短片分镜和视频的 Managed Agent。”                  |
 | 图片和视频生成   | “生成一张穿着太空服的猫站在火星上的图片，再把它制作成一段视频。”        |
+| 语音识别         | “把这段音频转写成文字，专有名词识别不准的话帮我加上热词再试。”          |
 | 用量与额度       | “查看最近的模型用量、免费额度和限流情况。”                              |
 | 模型选型         | “推荐一个适合图片理解和智能客服的模型。”                                |
 | 了解 Bailian CLI | “介绍一下 Bailian CLI 能帮我完成哪些任务，并根据我的需求推荐使用方式。” |

docs/agents/command-flag-change.md

@@ -19,6 +19,7 @@
   - 类型由 `ParsedFlags<typeof FLAGS>` 推导;避免手写 `flags.x as number` 这类断言
   - 单 flag 必填用 `required: true`;跨 flag / 值相关校验放 `validate`
   - 默认值 fallback 写在命令实现或 `Settings` 解析层,不要重复解析 env/config
+  - 需要在高风险确认前检查本地路径时,可用异步 `validate`;runtime 会在鉴权和确认前等待它完成。这里只允许本地只读检查,不写文件、不请求远端。非缺参的环境错误应抛出 `BailianError`,避免裸命令调用被当成缺参而仅显示 help。
 
 ### B. 鉴权 / 全局选项
 

docs/agents/telemetry-change.md

@@ -30,7 +30,7 @@
 | 命令声明          | 凭证 / 请求域                                       | 主要请求出口                                                                          | 后端埋点标识                                  | 前端埋点标识（AEM）                              |
 | ----------------- | --------------------------------------------------- | ------------------------------------------------------------------------------------- | --------------------------------------------- | ------------------------------------------------ |
 | `auth: "apiKey"`  | API Key；DashScope / OpenAI-compatible 模型域       | `Client.request/requestJson`、`McpClient`、Managed Agent instrumented fetch、上传策略 | 有：`User-Agent`、`x-dashscope-source-config` | 有：`pid=bailian-cli-node`、`authMethod=apiKey`  |
-| `auth: "console"` | Console access token；Bailian Console Gateway       | `callConsoleGateway()` → `/cli/api.json`                                              | 无                                            | 有：`pid=bailian-cli-node`、`authMethod=console` |
+| `auth: "console"` | Console access token；Bailian Console Gateway       | `callConsoleGateway()` → `/cli/api.json`                                              | 有：Console Gateway 后端直接标识              | 有：`pid=bailian-cli-node`、`authMethod=console` |
 | `auth: "openapi"` | AccessKey ID/Secret，可选 STS token；阿里云 OpenAPI | `Client.openApiJson()`                                                                | 有：`x-dashscope-source-config`               | 有：`pid=bailian-cli-node`、`authMethod=openapi` |
 | `auth: "none"`    | 无凭证域                                            | 本地逻辑或命令自行管理的登录/配置流程                                                 | 无                                            | 有：`pid=bailian-cli-node`、`authMethod=none`    |
 
@@ -41,6 +41,7 @@
 
 - Managed Agent 的 `User-Agent` 对所有 SDK 请求注入；`x-dashscope-source-config` 仅对阿里云 host 注入
 - DashScope 上传策略 `getPolicy` 只有 `x-dashscope-source-config`，没有显式 CLI `User-Agent`
+- Console Gateway 用户数据由后端直接区分，不依赖 `x-dashscope-source-config`；AEM 的 `authMethod=console` 是独立的命令侧统计
 - OpenAPI 的 ACS 签名头，以及 Console Gateway 的 `product`、`action`、`api` 是鉴权或路由字段，不计为埋点标识
 
 ### 2. 后端渠道参数

packages/cli/README.md

@@ -115,6 +115,7 @@ Once installed, just describe your task to your AI Agent — no need to assemble
 | ------------------------ | --------------------------------------------------------------------------------- |
 | Managed Agent            | "Create a Managed Agent that can generate short-film storyboards and videos."     |
 | Image & video generation | "Generate an image of a cat in a spacesuit on Mars, then turn it into a video."   |
+| Speech recognition       | "Transcribe this audio; if proper nouns are wrong, add hot words and try again."  |
 | Usage & quota            | "Show my recent model usage, free-tier quota, and rate limits."                   |
 | Model selection          | "Recommend a model for image understanding and customer support."                 |
 | About Bailian CLI        | "Tell me what Bailian CLI can do for me, and suggest how to use it for my needs." |

packages/cli/README.zh.md

@@ -114,6 +114,7 @@ irm https://bailian.aliyun.com/cli/install.ps1 | iex
 | ---------------- | ----------------------------------------------------------------------- |
 | Managed Agent    | “帮我创建一个能够生成短片分镜和视频的 Managed Agent。”                  |
 | 图片和视频生成   | “生成一张穿着太空服的猫站在火星上的图片，再把它制作成一段视频。”        |
+| 语音识别         | “把这段音频转写成文字，专有名词识别不准的话帮我加上热词再试。”          |
 | 用量与额度       | “查看最近的模型用量、免费额度和限流情况。”                              |
 | 模型选型         | “推荐一个适合图片理解和智能客服的模型。”                                |
 | 了解 Bailian CLI | “介绍一下 Bailian CLI 能帮我完成哪些任务，并根据我的需求推荐使用方式。” |

packages/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bailian-cli",
-  "version": "1.20.0",
+  "version": "1.23.0",
   "description": "CLI for Aliyun Model Studio (DashScope) AI Platform.",
   "keywords": [
     "agent",

packages/cli/src/commands.ts

@@ -70,6 +70,11 @@ import {
   searchWeb,
   speechSynthesize,
   speechRecognize,
+  speechVocabularyCreate,
+  speechVocabularyList,
+  speechVocabularyGet,
+  speechVocabularyUpdate,
+  speechVocabularyDelete,
   fileUpload,
   consoleCall,
   usageFree,
@@ -140,6 +145,18 @@ import {
   managedAgentPlan,
   managedAgentApply,
   managedAgentDestroy,
+  managedAgentPlayground,
+  managedAgentProjectInit,
+  managedAgentProjectValidate,
+  managedAgentProjectBuild,
+  managedAgentProjectPublish,
+  managedAgentProjectWorkbench,
+  managedAgentProjectVersionEnable,
+  managedAgentProjectVersionDisable,
+  managedAgentProjectVersionStatus,
+  managedAgentProjectVersionList,
+  managedAgentProjectVersionPreview,
+  managedAgentProjectVersionRestore,
   managedAgentStateList,
   managedAgentStateShow,
   managedAgentStateRm,
@@ -274,6 +291,11 @@ export const commands: Record<string, AnyCommand> = {
   "search web": searchWeb,
   "speech synthesize": speechSynthesize,
   "speech recognize": speechRecognize,
+  "speech vocabulary create": speechVocabularyCreate,
+  "speech vocabulary list": speechVocabularyList,
+  "speech vocabulary get": speechVocabularyGet,
+  "speech vocabulary update": speechVocabularyUpdate,
+  "speech vocabulary delete": speechVocabularyDelete,
   "file upload": fileUpload,
   "console call": consoleCall,
   "usage free": usageFree,
@@ -344,6 +366,18 @@ export const commands: Record<string, AnyCommand> = {
   "managed-agent plan": managedAgentPlan,
   "managed-agent apply": managedAgentApply,
   "managed-agent destroy": managedAgentDestroy,
+  "managed-agent playground": managedAgentPlayground,
+  "managed-agent project init": managedAgentProjectInit,
+  "managed-agent project validate": managedAgentProjectValidate,
+  "managed-agent project build": managedAgentProjectBuild,
+  "managed-agent project publish": managedAgentProjectPublish,
+  "managed-agent project workbench": managedAgentProjectWorkbench,
+  "managed-agent project version enable": managedAgentProjectVersionEnable,
+  "managed-agent project version disable": managedAgentProjectVersionDisable,
+  "managed-agent project version status": managedAgentProjectVersionStatus,
+  "managed-agent project version list": managedAgentProjectVersionList,
+  "managed-agent project version preview": managedAgentProjectVersionPreview,
+  "managed-agent project version restore": managedAgentProjectVersionRestore,
   "managed-agent state list": managedAgentStateList,
   "managed-agent state show": managedAgentStateShow,
   "managed-agent state rm": managedAgentStateRm,

packages/commands/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bailian-cli-commands",
-  "version": "1.20.0",
+  "version": "1.23.0",
   "description": "Command library for bailian-cli products (knowledge, memory, media, …). See https://www.npmjs.com/package/bailian-cli for usage.",
   "homepage": "https://bailian.console.aliyun.com/cli",
   "bugs": {
@@ -40,7 +40,7 @@
     "check": "vp check"
   },
   "dependencies": {
-    "@openagentpack/sdk": "0.5.0",
+    "@openagentpack/sdk": "0.7.1",
     "bailian-cli-core": "workspace:*",
     "bailian-cli-runtime": "workspace:*",
     "boxen": "catalog:",

packages/commands/src/commands/config/set.ts

@@ -12,9 +12,9 @@ export default defineCommand({
       valueHint: "<key>",
       description: {
         "en-US":
-          "Config key (language, base_url, output, output_dir, timeout, api_key, api_key_capabilities, access_token, access_key_id, access_key_secret, security_token, default_*_model, workspace_id)",
+          "Config key (language, base_url, output, output_dir, timeout, watermark, api_key, api_key_capabilities, access_token, access_key_id, access_key_secret, security_token, default_*_model, workspace_id)",
         "zh-CN":
-          "配置项名称（language、base_url、output、output_dir、timeout、api_key、api_key_capabilities、access_token、access_key_id、access_key_secret、security_token、default_*_model、workspace_id）",
+          "配置项名称（language、base_url、output、output_dir、timeout、watermark、api_key、api_key_capabilities、access_token、access_key_id、access_key_secret、security_token、default_*_model、workspace_id）",
       },
       required: true,
     },
@@ -29,6 +29,7 @@ export default defineCommand({
     "--key language --value zh-CN",
     "--key output --value json",
     "--key timeout --value 600",
+    "--key watermark --value false",
     "--key base_url --value https://dashscope.aliyuncs.com",
     "--config company-plan --key api-key-capabilities --value text.chat,image.generate",
   ],

packages/commands/src/commands/config/shared.ts

@@ -3,6 +3,7 @@ import {
   ExitCode,
   isApiKeyCapability,
   normalizeModelBaseUrl,
+  parseBooleanValue,
   SUPPORTED_LANGUAGES,
 } from "bailian-cli-core";
 
@@ -13,6 +14,7 @@ export const VALID_KEYS = [
   "output",
   "output_dir",
   "timeout",
+  "watermark",
   "api_key",
   "access_token",
   "access_key_id",
@@ -62,7 +64,7 @@ export const UI_ENUM_KEYS: Record<string, string[]> = {
 };
 
 // Keys the UI renders as a true/false dropdown and stores as a boolean.
-export const UI_BOOLEAN_KEYS = new Set<string>(["telemetry"]);
+export const UI_BOOLEAN_KEYS = new Set<string>(["telemetry", "watermark"]);
 
 // Default model each `default_*_model` key falls back to when left unset. These
 // mirror the inline `|| "<model>"` fallbacks in the generation commands
@@ -164,7 +166,10 @@ export function resolveKey(key: string): string {
  * Validate a single config entry and coerce its value to the stored type.
  * Throws BailianError(USAGE) for unknown keys or invalid values.
  */
-export function validateAndCoerce(key: string, value: string): string | number | string[] {
+export function validateAndCoerce(
+  key: string,
+  value: string,
+): string | number | boolean | string[] {
   const resolvedKey = resolveKey(key);
 
   if (!(VALID_KEYS as readonly string[]).includes(resolvedKey)) {
@@ -201,6 +206,8 @@ export function validateAndCoerce(key: string, value: string): string | number |
 
   if (resolvedKey === "base_url") return normalizeModelBaseUrl(value);
 
+  if (resolvedKey === "watermark") return parseBooleanValue(value, "watermark");
+
   if (resolvedKey === "api_key_capabilities") {
     let rawCapabilities: unknown;
     if (value.trim().startsWith("[")) {

packages/commands/src/commands/config/show.ts

@@ -18,6 +18,7 @@ export default defineCommand({
       base_url: client.baseUrl,
       output: settings.output,
       timeout: settings.timeout,
+      watermark: settings.watermark,
       config: settings.configName ?? "default",
       config_file: store.path,
     };

packages/commands/src/commands/image/edit.ts

@@ -207,7 +207,7 @@ export default defineCommand({
       "prompt-extend",
     );
 
-    const watermark = resolveWatermark(flags.watermark);
+    const watermark = resolveWatermark(flags.watermark, settings.watermark);
 
     const parameters: NonNullable<DashScopeImageRequest["parameters"]> = {
       size: resolveImageSize(flags.size, route.sizeProfile),

packages/commands/src/commands/image/generate.ts

@@ -185,7 +185,7 @@ export default defineCommand({
       "prompt-extend",
     );
 
-    const watermark = resolveWatermark(flags.watermark);
+    const watermark = resolveWatermark(flags.watermark, settings.watermark);
 
     const parameters: NonNullable<DashScopeImageRequest["parameters"]> = {
       size,

packages/commands/src/commands/managed-agent/_engine/config-loader.ts

@@ -30,9 +30,11 @@ export const BAILIAN_PROVIDER = "bailian";
 
 interface AgentConfigOptions {
   resolveEnv?: boolean;
+  environment?: Record<string, string>;
   projectName?: string;
   statePath?: string;
   credentials?: CredentialScope;
+  overrideBailianBaseUrl?: boolean;
 }
 
 /**
@@ -56,7 +58,9 @@ export async function resolveAgentProjectConfig(
   prepareProviderEnv();
   const resolved = await resolveProjectConfig(filePath, options);
   normalizeInterpolatedProviderBlocks(resolved.config.providers);
-  injectProviderCredentials(resolved.config.providers, host);
+  injectProviderCredentials(resolved.config.providers, host, {
+    overrideBaseUrl: options.overrideBailianBaseUrl,
+  });
   scrubCredentialEnv();
   assertBailianOnlyProviders(resolved.config.providers);
   if ((options.credentials ?? "all") !== "none") {

packages/commands/src/commands/managed-agent/_engine/credentials.ts

@@ -99,9 +99,10 @@ export function prepareProviderEnv(): void {
  * Override the bailian provider block with bl's authStage-resolved credential, so
  * the bailian API key is authoritatively the CLI auth chain's — never a config
  * file bare-read or a stale env value. `api_key` is replaced unconditionally
- * when a credential resolved; `base_url` / `workspace_id` are filled only when
- * the block references them and the interpolated value is empty (a literal in
- * agents.yaml is respected).
+ * when a credential resolved. `base_url` is normally filled only when empty so
+ * a literal in agents.yaml remains supported; directory projects pass
+ * `overrideBaseUrl` so their connection always follows the Bailian CLI auth
+ * chain. `workspace_id` is filled only when empty.
  *
  * `base_url` carries {@link AGENTSTUDIO_API_PATH} because the SDK appends resource
  * paths onto it verbatim; a value already ending in the suffix is left as-is.
@@ -114,6 +115,7 @@ export function prepareProviderEnv(): void {
 export function injectProviderCredentials(
   providers: Record<string, unknown>,
   host: CredentialHost,
+  options: { overrideBaseUrl?: boolean } = {},
 ): void {
   const bailian = providers.bailian;
   if (!bailian || typeof bailian !== "object") return;
@@ -121,7 +123,7 @@ export function injectProviderCredentials(
 
   const cred = host.client.exportApiCredential();
   if (cred) block.api_key = cred.token;
-  if ("base_url" in block && !block.base_url) {
+  if ("base_url" in block && (options.overrideBaseUrl || !block.base_url)) {
     // Defensive normalization: the auth chain already normalizes base_url to
     // an origin, but never let a trailing slash produce "//api/v1/agentstudio".
     const origin = host.client.baseUrl.replace(/\/+$/, "");

packages/commands/src/commands/managed-agent/_engine/playground-launcher.ts

@@ -0,0 +1,432 @@
+import { execFile, spawn, type ChildProcess } from "node:child_process";
+import { createHash, randomBytes } from "node:crypto";
+import { existsSync, readFileSync } from "node:fs";
+import { createRequire } from "node:module";
+import { dirname, resolve } from "node:path";
+import { promisify } from "node:util";
+import { BailianError, type Client, ExitCode, type Settings } from "bailian-cli-core";
+import { emitBare } from "bailian-cli-runtime";
+
+const PLAYGROUND_PACKAGE = "@openagentpack/playground";
+const execFileAsync = promisify(execFile);
+const DEFAULT_PORT = 4848;
+const PLAYGROUND_URL_PATTERN = /running at http:\/\/localhost:(\d+)/i;
+
+export interface PlaygroundLaunchOptions {
+  port?: number;
+  open: boolean;
+  file?: string;
+  project?: string;
+  agent?: string;
+  surface: "preview" | "workbench";
+  client: Client;
+  settings: Settings;
+}
+
+interface Launcher {
+  command: string;
+  args: string[];
+  version?: string;
+  fetched: boolean;
+}
+
+interface ExistingPlayground {
+  version: string;
+  pid: number;
+  projectId?: string;
+}
+
+interface PlaygroundProjectSummary {
+  status?: string;
+  agents?: Array<{ agent?: { id?: string } }>;
+}
+
+export interface PlaygroundBrowserTarget {
+  url: string;
+  warning?: string;
+}
+
+export async function launchManagedAgentPlayground(
+  options: PlaygroundLaunchOptions,
+): Promise<void> {
+  assertSupportedNodeVersion();
+  const port = options.port ?? DEFAULT_PORT;
+  if (!Number.isInteger(port) || port <= 0 || port > 65_535) {
+    throw new BailianError(`Invalid --port '${port}'.`, ExitCode.USAGE);
+  }
+  const sourcePath = resolve(
+    options.surface === "workbench" ? (options.project ?? ".") : (options.file ?? "agents.yaml"),
+  );
+  const projectId = createHash("sha256").update(sourcePath).digest("hex").slice(0, 16);
+  const launcher = await resolveLauncher();
+  const existing = await probeExistingPlayground(port);
+  if (existing) {
+    const reusable =
+      existing.projectId === projectId &&
+      (launcher.version === undefined || existing.version === launcher.version);
+    if (reusable) {
+      emitBare(`Workbench already running at http://localhost:${port} (pid ${existing.pid}).`);
+      await openPlaygroundSurface(port, options.surface, options.agent, options.open);
+      return;
+    }
+    const released = await replaceExistingPlayground(existing, port);
+    if (!released) {
+      throw new BailianError(
+        `Could not stop the existing Workbench process (pid ${existing.pid}) on port ${port}.`,
+        ExitCode.GENERAL,
+        "Stop it manually or choose another --port.",
+      );
+    }
+  }
+
+  const environment = buildPlaygroundEnvironment(options, port, sourcePath);
+  if (launcher.fetched) {
+    emitBare(`Fetching ${PLAYGROUND_PACKAGE} (first run may take a moment)...`);
+  }
+  const child = spawn(launcher.command, launcher.args, {
+    env: environment,
+    stdio: ["inherit", "pipe", "inherit"],
+  });
+  const removeSignalForwarding = forwardSignals(child);
+  try {
+    const readyPort = await waitForPlaygroundReady(child, port, 30_000, projectId);
+    if (readyPort === null) {
+      throw new BailianError(
+        `Workbench did not become ready in time. Check the logs above, then open http://localhost:${port}.`,
+        ExitCode.GENERAL,
+      );
+    }
+    emitBare(`Workbench ready at http://localhost:${readyPort}`);
+    await openPlaygroundSurface(readyPort, options.surface, options.agent, options.open);
+    const exitCode = await waitForChildExit(child);
+    if (exitCode !== 0) {
+      throw new BailianError(`Workbench exited with code ${exitCode}.`, ExitCode.GENERAL);
+    }
+  } finally {
+    removeSignalForwarding();
+  }
+}
+
+export function playgroundBrowserTargetFromSummary(
+  baseUrl: string,
+  summary: PlaygroundProjectSummary,
+  requestedAgent?: string,
+): PlaygroundBrowserTarget {
+  if (summary.status !== "valid") return { url: baseUrl };
+  const agentIds = (summary.agents ?? [])
+    .map((entry) => entry.agent?.id?.trim())
+    .filter((agentId): agentId is string => Boolean(agentId));
+  const requested = requestedAgent?.trim();
+  if (requested) {
+    if (agentIds.includes(requested)) {
+      return { url: `${baseUrl}/agents/${encodeURIComponent(requested)}/preview` };
+    }
+    return {
+      url: baseUrl,
+      warning: `Agent '${requested}' was not found. Opening the project Workbench instead.`,
+    };
+  }
+  if (agentIds.length === 1) {
+    return { url: `${baseUrl}/agents/${encodeURIComponent(agentIds[0]!)}/preview` };
+  }
+  if (agentIds.length > 1) {
+    return {
+      url: baseUrl,
+      warning:
+        "This project declares multiple Agents. Opening the Workbench; rerun with --agent <id> for Preview.",
+    };
+  }
+  return { url: baseUrl };
+}
+
+function assertSupportedNodeVersion(): void {
+  const majorVersion = Number(process.versions.node.split(".")[0]);
+  if (Number.isFinite(majorVersion) && majorVersion >= 22) return;
+  throw new BailianError(
+    "Managed Agent Workbench requires Node.js 22 or later.",
+    ExitCode.USAGE,
+    "Upgrade Node.js for Workbench; other Bailian CLI commands continue to support Node.js 18.17+.",
+  );
+}
+
+export async function resolveLauncher(): Promise<Launcher> {
+  const explicit =
+    process.env.BAILIAN_MANAGED_AGENT_PLAYGROUND_BIN?.trim() ||
+    process.env.AGENTS_PLAYGROUND_BIN?.trim();
+  if (explicit) {
+    if (!existsSync(explicit)) {
+      throw new BailianError(
+        `Configured Workbench binary does not exist: ${explicit}`,
+        ExitCode.USAGE,
+      );
+    }
+    return { command: process.execPath, args: [explicit], fetched: false };
+  }
+
+  const requestedVersion = process.env.BAILIAN_MANAGED_AGENT_PLAYGROUND_VERSION?.trim() || "latest";
+  const { stdout } = await execFileAsync(
+    "npm",
+    [
+      "view",
+      `${PLAYGROUND_PACKAGE}@${requestedVersion}`,
+      "version",
+      "--json",
+      "--prefer-online",
+      "--fetch-retries=0",
+      "--fetch-timeout=10000",
+    ],
+    { timeout: 15_000, maxBuffer: 1024 * 1024 },
+  );
+  let resolvedVersion: unknown;
+  try {
+    resolvedVersion = JSON.parse(stdout);
+  } catch {
+    resolvedVersion = undefined;
+  }
+  if (
+    typeof resolvedVersion !== "string" ||
+    !/^\d+\.\d+\.\d+(?:-[\da-zA-Z.-]+)?(?:\+[\da-zA-Z.-]+)?$/.test(resolvedVersion)
+  ) {
+    throw new BailianError(
+      "npm did not return a single valid Playground version. Specify an exact version or dist-tag. / npm 未返回唯一有效的 Playground 版本号。请指定精确版本或 dist-tag。",
+      ExitCode.GENERAL,
+    );
+  }
+
+  const installed = resolveInstalledPlayground();
+  if (installed?.version === resolvedVersion) return installed;
+
+  const monorepoBinary = findLocalPlaygroundBin(process.cwd());
+  if (
+    monorepoBinary &&
+    readPlaygroundVersion(resolve(dirname(monorepoBinary), "../../package.json")) ===
+      resolvedVersion
+  ) {
+    return {
+      command: process.execPath,
+      args: [monorepoBinary],
+      version: resolvedVersion,
+      fetched: false,
+    };
+  }
+
+  return {
+    command: "npx",
+    args: ["-y", `${PLAYGROUND_PACKAGE}@${resolvedVersion}`],
+    version: resolvedVersion,
+    fetched: true,
+  };
+}
+
+function readPlaygroundVersion(manifestPath: string): string | undefined {
+  try {
+    return (JSON.parse(readFileSync(manifestPath, "utf8")) as { version?: string }).version;
+  } catch {
+    return undefined;
+  }
+}
+
+function resolveInstalledPlayground(): Launcher | undefined {
+  try {
+    const require = createRequire(import.meta.url);
+    const packageJsonPath = require.resolve(`${PLAYGROUND_PACKAGE}/package.json`);
+    const manifest = JSON.parse(readFileSync(packageJsonPath, "utf8")) as {
+      version?: string;
+      bin?: string | Record<string, string>;
+    };
+    const relativeBinary =
+      typeof manifest.bin === "string" ? manifest.bin : manifest.bin?.["agents-playground"];
+    if (!relativeBinary) return undefined;
+    const binaryPath = resolve(dirname(packageJsonPath), relativeBinary);
+    if (!existsSync(binaryPath)) return undefined;
+    return {
+      command: process.execPath,
+      args: [binaryPath],
+      version: manifest.version,
+      fetched: false,
+    };
+  } catch {
+    return undefined;
+  }
+}
+
+function findLocalPlaygroundBin(startDirectory: string): string | undefined {
+  let directory = startDirectory;
+  for (let depth = 0; depth < 10; depth += 1) {
+    const candidate = resolve(directory, "packages/playground/dist/bin/playground.js");
+    if (existsSync(candidate)) return candidate;
+    const parent = dirname(directory);
+    if (parent === directory) break;
+    directory = parent;
+  }
+  return undefined;
+}
+
+function buildPlaygroundEnvironment(
+  options: PlaygroundLaunchOptions,
+  port: number,
+  sourcePath: string,
+): NodeJS.ProcessEnv {
+  const credential = options.client.exportApiCredential();
+  const environment: NodeJS.ProcessEnv = {
+    ...process.env,
+    PORT: String(port),
+    AGENTS_PLAYGROUND_TOKEN: randomBytes(32).toString("hex"),
+  };
+  if (options.surface === "workbench") {
+    delete environment.AGENTS_CONFIG_PATH;
+    environment.AGENTS_PROJECT_ROOT = sourcePath;
+  } else {
+    delete environment.AGENTS_PROJECT_ROOT;
+    environment.AGENTS_CONFIG_PATH = sourcePath;
+  }
+  if (credential) environment.DASHSCOPE_API_KEY = credential.token;
+  const baseUrl = options.client.baseUrl.replace(/\/+$/, "");
+  environment.BAILIAN_BASE_URL = baseUrl.endsWith("/api/v1/agentstudio")
+    ? baseUrl
+    : `${baseUrl}/api/v1/agentstudio`;
+  if (options.settings.workspaceId) {
+    environment.BAILIAN_WORKSPACE_ID = options.settings.workspaceId;
+  }
+  return environment;
+}
+
+async function waitForPlaygroundReady(
+  child: ChildProcess,
+  fallbackPort: number,
+  timeoutMs: number,
+  expectedProjectId: string,
+): Promise<number | null> {
+  let port = fallbackPort;
+  let outputBuffer = "";
+  child.stdout?.on("data", (chunk: Buffer | string) => {
+    process.stdout.write(chunk);
+    outputBuffer += chunk.toString();
+    const match = outputBuffer.match(PLAYGROUND_URL_PATTERN);
+    if (match?.[1]) port = Number(match[1]);
+  });
+
+  const deadline = Date.now() + timeoutMs;
+  while (Date.now() <= deadline) {
+    if (child.exitCode !== null) return null;
+    try {
+      const response = await fetch(`http://localhost:${port}/health`, {
+        signal: AbortSignal.timeout(1_000),
+      });
+      const body = response.ok
+        ? ((await response.json()) as { playground?: { project_id?: string } })
+        : undefined;
+      if (body?.playground?.project_id === expectedProjectId) return port;
+    } catch {
+      // Not ready yet.
+    }
+    await new Promise<void>((resolveWait) => setTimeout(resolveWait, 300));
+  }
+  return null;
+}
+
+async function probeExistingPlayground(port: number): Promise<ExistingPlayground | null> {
+  try {
+    const response = await fetch(`http://localhost:${port}/health`, {
+      signal: AbortSignal.timeout(2_000),
+    });
+    if (!response.ok) return null;
+    const body = (await response.json()) as {
+      playground?: { version?: string; pid?: number; project_id?: string };
+    };
+    if (!body.playground?.pid) return null;
+    return {
+      version: body.playground.version ?? "unknown",
+      pid: body.playground.pid,
+      projectId: body.playground.project_id,
+    };
+  } catch {
+    return null;
+  }
+}
+
+async function replaceExistingPlayground(
+  existing: ExistingPlayground,
+  port: number,
+): Promise<boolean> {
+  emitBare(`Replacing Workbench v${existing.version} (pid ${existing.pid}) on port ${port}...`);
+  try {
+    process.kill(existing.pid, "SIGTERM");
+  } catch {
+    return true;
+  }
+  for (let attempt = 0; attempt < 30; attempt += 1) {
+    await new Promise<void>((resolveWait) => setTimeout(resolveWait, 100));
+    if (!(await probeExistingPlayground(port))) return true;
+  }
+  return false;
+}
+
+async function openPlaygroundSurface(
+  port: number,
+  surface: "preview" | "workbench",
+  requestedAgent: string | undefined,
+  shouldOpen: boolean,
+): Promise<void> {
+  if (!shouldOpen) return;
+  const target =
+    surface === "workbench"
+      ? { url: `http://localhost:${port}` }
+      : await resolvePlaygroundBrowserTarget(port, requestedAgent);
+  if (target.warning) emitBare(`Warning: ${target.warning}`);
+  openBrowser(target.url);
+}
+
+async function resolvePlaygroundBrowserTarget(
+  port: number,
+  requestedAgent?: string,
+): Promise<PlaygroundBrowserTarget> {
+  const baseUrl = `http://localhost:${port}`;
+  try {
+    const response = await fetch(`${baseUrl}/api/project`, {
+      signal: AbortSignal.timeout(3_000),
+    });
+    if (!response.ok) return { url: baseUrl };
+    return playgroundBrowserTargetFromSummary(
+      baseUrl,
+      (await response.json()) as PlaygroundProjectSummary,
+      requestedAgent,
+    );
+  } catch {
+    return { url: baseUrl };
+  }
+}
+
+function openBrowser(url: string): void {
+  const command =
+    process.platform === "darwin" ? "open" : process.platform === "win32" ? "start" : "xdg-open";
+  const args = process.platform === "win32" ? ["", url] : [url];
+  try {
+    spawn(command, args, {
+      stdio: "ignore",
+      detached: true,
+      shell: process.platform === "win32",
+    }).unref();
+  } catch {
+    emitBare(`Could not open a browser automatically. Visit ${url}`);
+  }
+}
+
+function forwardSignals(child: ChildProcess): () => void {
+  const forwardInterrupt = () => child.kill("SIGINT");
+  const forwardTerminate = () => child.kill("SIGTERM");
+  process.on("SIGINT", forwardInterrupt);
+  process.on("SIGTERM", forwardTerminate);
+  return () => {
+    process.off("SIGINT", forwardInterrupt);
+    process.off("SIGTERM", forwardTerminate);
+  };
+}
+
+function waitForChildExit(child: ChildProcess): Promise<number> {
+  if (child.exitCode !== null) return Promise.resolve(child.exitCode);
+  return new Promise((resolveExit, rejectExit) => {
+    child.once("exit", (exitCode) => resolveExit(exitCode ?? 0));
+    child.once("error", rejectExit);
+  });
+}

packages/commands/src/commands/managed-agent/init.ts

@@ -1,5 +1,6 @@
 import { existsSync } from "node:fs";
 import { readFile, writeFile } from "node:fs/promises";
+import { resolve } from "node:path";
 import {
   BailianError,
   defineCommand,
@@ -12,6 +13,7 @@ import { emitBare, emitResult } from "bailian-cli-runtime";
 const GITIGNORE_ADDITIONS = `
 # agents
 agents.state.json
+.openagentpack/versions/
 .env
 `;
 
@@ -71,8 +73,8 @@ const INIT_FLAGS = {
 
 export default defineCommand({
   description: {
-    "en-US": "Create a new agents.yaml template",
-    "zh-CN": "创建新的 agents.yaml 模板",
+    "en-US": "Create an agents.yaml template",
+    "zh-CN": "创建 agents.yaml 模板",
   },
   auth: "none",
   usageArgs: "[--agent-name <name>] [--file <path>] [--force]",
@@ -86,7 +88,7 @@ export default defineCommand({
 
     if (existsSync(file) && !flags.force) {
       throw new BailianError(
-        `${file} already exists.`,
+        `${resolve(file)} already exists.`,
         ExitCode.USAGE,
         "Pass --force to overwrite.",
       );
@@ -127,7 +129,7 @@ export default defineCommand({
     if (format === "json") {
       emitResult({ created: file, provider: "bailian", agent: agentName }, format);
     } else {
-      emitBare(`Created ${file}`);
+      emitBare(`Created ${resolve(file)}`);
       emitBare(
         "Credentials: run `bl auth login --api-key <key> --base-url <url>`, or set DASHSCOPE_API_KEY / BAILIAN_BASE_URL.",
       );

packages/commands/src/commands/managed-agent/project.ts

@@ -0,0 +1,468 @@
+import { stat } from "node:fs/promises";
+import { join } from "node:path";
+import {
+  BailianError,
+  defineCommand,
+  detectOutputFormat,
+  ExitCode,
+  type FlagsDef,
+} from "bailian-cli-core";
+import { emitBare, emitResult } from "bailian-cli-runtime";
+import {
+  commitProjectBuild,
+  createDirectoryWorkspaceVersionService,
+  executeProjectPublish,
+  initializeDirectoryProject,
+  planProjectPublish,
+  previewProjectBuild,
+  type ProjectBuildResolver,
+  resolveDirectoryProjectRoot,
+  validateDirectoryProject,
+} from "@openagentpack/sdk/project-workspace";
+import { CREDENTIALS_NOTE, resolveAgentProjectConfig } from "./_engine/config-loader.ts";
+import { withStdoutProtected } from "./_engine/console-capture.ts";
+import { withAgentErrors } from "./_engine/errors.ts";
+import { renderAgentFeedback } from "./_engine/feedback.ts";
+import { launchManagedAgentPlayground } from "./_engine/playground-launcher.ts";
+import { installSdkTransport } from "./_engine/transport.ts";
+import { formatResourceLabel } from "./_engine/address-utils.ts";
+
+const PROJECT_FLAG = {
+  project: {
+    type: "string",
+    valueHint: "<directory>",
+    description: {
+      "en-US": "Directory project root (default: current directory)",
+      "zh-CN": "目录项目根路径（默认：当前目录）",
+    },
+  },
+} satisfies FlagsDef;
+
+const JSON_FORMAT = "json" as const;
+
+export const managedAgentProjectInit = defineCommand({
+  description: {
+    "en-US": "Create a directory project or convert the local agents.yaml",
+    "zh-CN": "创建目录项目，或转换当前 agents.yaml",
+  },
+  auth: "none",
+  usageArgs: "[--project <directory>]",
+  flags: {
+    project: {
+      ...PROJECT_FLAG.project,
+      description: {
+        "en-US": "Directory project root (default: ./managed-agent under the current directory)",
+        "zh-CN": "目录项目根路径（默认：当前目录下的 ./managed-agent）",
+      },
+    },
+  },
+  exampleArgs: ["", "--project ./my-agent", "--project ."],
+  notes: [
+    {
+      "en-US":
+        "Without --project, creates a managed-agent/ subdirectory. Enter it before running other project commands. Use --project . to initialize in place or convert the current agents.yaml; existing project files are not overwritten.",
+      "zh-CN":
+        "不传 --project 时创建 managed-agent/ 子目录；后续项目操作请先进入该目录。使用 --project . 可在当前目录初始化或转换 agents.yaml；不会覆盖已有项目文件。",
+    },
+    {
+      "en-US":
+        "New projects include Skill, File, Vault, and Environment examples under each resource directory's _examples/. They are not referenced by agent.json and are excluded from Build/Publish. Copy an example outside _examples/ to enable it, then configure its Agent reference.",
+      "zh-CN":
+        "新项目在四类资源目录的 _examples/ 下生成配置示例，不写入 agent.json 引用，也不参与 Build/Publish。需要使用时，将示例复制到 _examples/ 外，再配置 Agent 引用。",
+    },
+  ],
+  async run(ctx) {
+    const projectRoot = ctx.flags.project ?? "./managed-agent";
+    if (ctx.settings.dryRun) {
+      emitResult(
+        {
+          would_initialize_project: projectRoot,
+        },
+        detectOutputFormat(ctx.settings.output),
+      );
+      return;
+    }
+    const result = await initializeDirectoryProject({ projectRoot });
+    emitResult(result, detectOutputFormat(ctx.settings.output));
+  },
+});
+
+export const managedAgentProjectValidate = defineCommand({
+  description: { "en-US": "Validate a directory Agent project", "zh-CN": "校验目录式 Agent 项目" },
+  auth: "none",
+  usageArgs: "[--project <directory>]",
+  flags: PROJECT_FLAG,
+  exampleArgs: ["", "--project ./my-agent"],
+  async run(ctx) {
+    const result = await validateDirectoryProject(ctx.flags.project ?? ".");
+    const format = detectOutputFormat(ctx.settings.output);
+    if (format === JSON_FORMAT) emitResult(safeInspection(result), format);
+    else {
+      for (const diagnostic of [...result.diagnostics, ...result.warnings]) {
+        emitBare(`${diagnostic.severity}: ${diagnostic.code}: ${diagnostic.message}`);
+      }
+      if (!result.diagnostics.some((diagnostic) => diagnostic.severity === "error")) {
+        emitBare(`Project is valid (${result.project_revision.slice(0, 12)}).`);
+      }
+    }
+    if (result.diagnostics.some((diagnostic) => diagnostic.severity === "error")) {
+      throw new BailianError("Directory project validation failed.", ExitCode.GENERAL);
+    }
+  },
+});
+
+export const managedAgentProjectBuild = defineCommand({
+  description: {
+    "en-US": "Organize directory source and generate the immutable Publish Build",
+    "zh-CN": "整理目录源文件并生成不可变的发布 Build",
+  },
+  auth: "none",
+  notes: [
+    {
+      "en-US":
+        "Build writes local project files without confirmation, including inferred resource associations and migration of plaintext Vault secrets into .env. Use --dry-run to preview without writing. Publish still requires explicit confirmation before remote changes.",
+      "zh-CN":
+        "Build 无需确认即可写入本地项目文件，包括推断的资源关联及将 Vault 明文密钥移入 .env。使用 --dry-run 可只预览不写入。Publish 变更远端资源前仍需显式确认。",
+    },
+  ],
+  usageArgs: "[--project <directory>]",
+  flags: PROJECT_FLAG,
+  exampleArgs: ["", "--dry-run", "--project ./my-agent"],
+  async validate(flags) {
+    await withAgentErrors(async () => {
+      const root = await resolveDirectoryProjectRoot(flags.project ?? ".");
+      const metadata = await stat(join(root, "project.json")).catch((error: unknown) => {
+        if ((error as NodeJS.ErrnoException).code === "ENOENT") return null;
+        throw error;
+      });
+      if (!metadata?.isFile()) {
+        throw new BailianError(
+          `Not a project root: ${root} (project.json is missing).`,
+          ExitCode.USAGE,
+          "Run from the directory containing project.json, or pass --project <directory>.",
+        );
+      }
+    });
+    return undefined;
+  },
+  async run(ctx) {
+    const root = ctx.flags.project ?? ".";
+    const preview = await previewProjectBuild(root);
+    const format = detectOutputFormat(ctx.settings.output);
+    if (ctx.settings.dryRun) {
+      emitResult(safeInspection(preview), format);
+      return;
+    }
+    if (!preview.can_build)
+      throw new BailianError(
+        preview.diagnostics.find((diagnostic) => diagnostic.severity === "error")?.message ??
+          "Directory project is invalid and cannot be built.",
+        ExitCode.GENERAL,
+      );
+    const built = await commitProjectBuild({
+      projectRoot: root,
+      baseRevision: preview.project_revision,
+    });
+    emitResult(
+      { manifest: built.manifest, organization_moves: preview.organization_moves },
+      format,
+    );
+  },
+});
+
+export const managedAgentProjectPublish = defineCommand({
+  description: {
+    "en-US": "Publish the current directory-project Build and record a version",
+    "zh-CN": "发布当前目录项目 Build 并记录版本",
+  },
+  auth: "apiKey",
+  risk: {
+    level: "high",
+    message: {
+      "en-US":
+        "This publishes the current directory-project Build and may create, update, or delete remote managed Agent resources.",
+      "zh-CN": "该操作会发布当前目录项目 Build，可能创建、更新或删除远端托管 Agent 资源。",
+    },
+  },
+  usageArgs: "[--project <directory>] [--no-refresh] [--concurrency <n>]",
+  flags: {
+    ...PROJECT_FLAG,
+    noRefresh: {
+      type: "switch",
+      description: {
+        "en-US": "Skip remote refresh before planning",
+        "zh-CN": "规划前跳过远端刷新",
+      },
+    },
+    concurrency: {
+      type: "number",
+      valueHint: "<n>",
+      description: {
+        "en-US": "Maximum parallel resource operations",
+        "zh-CN": "最大并行资源操作数",
+      },
+    },
+  },
+  exampleArgs: ["--yes", "--project ./my-agent --yes"],
+  notes: CREDENTIALS_NOTE,
+  async run(ctx) {
+    installSdkTransport(ctx);
+    const root = ctx.flags.project ?? ".";
+    const resolveBuild: ProjectBuildResolver = async (buildPath, options) =>
+      (await resolveAgentProjectConfig(ctx, buildPath, {
+        environment: options?.environment,
+        overrideBailianBaseUrl: true,
+      })) as unknown as Awaited<ReturnType<ProjectBuildResolver>>;
+    const planned = await withAgentErrors(() =>
+      withStdoutProtected(() =>
+        planProjectPublish(root, {
+          provider: "bailian",
+          refresh: !ctx.flags.noRefresh,
+          quiet: true,
+          onFeedback: renderAgentFeedback,
+          resolveBuild,
+        }),
+      ),
+    );
+    const actions = planned.planned.plan.actions.filter((action) => action.action !== "no-op");
+    if (ctx.settings.dryRun) {
+      emitResult(
+        {
+          project_revision: planned.project_revision,
+          build_manifest: planned.build_manifest,
+          plan: planned.planned.plan,
+        },
+        detectOutputFormat(ctx.settings.output),
+      );
+      return;
+    }
+    for (const action of actions) {
+      const marker = action.action === "create" ? "+" : action.action === "update" ? "~" : "-";
+      emitBare(`${marker} ${formatResourceLabel(action.address)}`);
+    }
+    const result = await withAgentErrors(() =>
+      withStdoutProtected(() =>
+        executeProjectPublish({
+          projectRoot: planned.project_root,
+          expectedProjectRevision: planned.project_revision,
+          expectedYamlHash: planned.build_manifest.yaml_hash,
+          expectedPlanFingerprint: planned.plan_fingerprint,
+          provider: "bailian",
+          refresh: !ctx.flags.noRefresh,
+          concurrency: ctx.flags.concurrency,
+          policy: "force",
+          onFeedback: renderAgentFeedback,
+          resolveBuild,
+        }),
+      ),
+    );
+    emitResult(result, detectOutputFormat(ctx.settings.output));
+  },
+});
+
+export const managedAgentProjectWorkbench = defineCommand({
+  description: {
+    "en-US": "Launch the directory project Workbench",
+    "zh-CN": "启动目录项目 Workbench",
+  },
+  auth: "apiKey",
+  usageArgs: "[--project <directory>] [--port <n>] [--no-open]",
+  flags: {
+    ...PROJECT_FLAG,
+    port: {
+      type: "number",
+      valueHint: "<n>",
+      description: { "en-US": "Local port (default: 4848)", "zh-CN": "本地端口（默认：4848）" },
+    },
+    noOpen: {
+      type: "switch",
+      description: { "en-US": "Do not open a browser", "zh-CN": "不自动打开浏览器" },
+    },
+  },
+  exampleArgs: ["", "--project ./my-agent --no-open"],
+  notes: CREDENTIALS_NOTE,
+  async run(ctx) {
+    const root = ctx.flags.project ?? ".";
+    if (ctx.settings.dryRun) {
+      emitResult(
+        { would_launch: "workbench", project_root: root, port: ctx.flags.port ?? 4848 },
+        detectOutputFormat(ctx.settings.output),
+      );
+      return;
+    }
+    await launchManagedAgentPlayground({
+      project: root,
+      port: ctx.flags.port ?? 4848,
+      open: !ctx.flags.noOpen,
+      surface: "workbench",
+      client: ctx.client,
+      settings: ctx.settings,
+    });
+  },
+});
+
+export const managedAgentProjectVersionEnable = versionToggleCommand(true);
+export const managedAgentProjectVersionDisable = versionToggleCommand(false);
+
+function versionToggleCommand(enabled: boolean) {
+  return defineCommand({
+    description: enabled
+      ? { "en-US": "Enable directory project versions", "zh-CN": "启用目录项目版本管理" }
+      : { "en-US": "Disable directory project versions", "zh-CN": "关闭目录项目版本管理" },
+    auth: "none",
+    usageArgs: "[--project <directory>]",
+    flags: PROJECT_FLAG,
+    exampleArgs: ["", "--project ./my-agent"],
+    async run(ctx) {
+      const service = createDirectoryWorkspaceVersionService(ctx.flags.project ?? ".");
+      if (ctx.settings.dryRun) {
+        emitResult(
+          { would_set_enabled: enabled, status: await service.status() },
+          detectOutputFormat(ctx.settings.output),
+        );
+        return;
+      }
+      const result = enabled
+        ? await service.enable("Enable project versions")
+        : await service.disable();
+      emitResult(result, detectOutputFormat(ctx.settings.output));
+    },
+  });
+}
+
+export const managedAgentProjectVersionStatus = defineCommand({
+  description: {
+    "en-US": "Show directory project version status",
+    "zh-CN": "显示目录项目版本状态",
+  },
+  auth: "none",
+  usageArgs: "[--project <directory>]",
+  flags: PROJECT_FLAG,
+  exampleArgs: ["", "--project ./my-agent --output json"],
+  async run(ctx) {
+    emitResult(
+      await createDirectoryWorkspaceVersionService(ctx.flags.project ?? ".").status(),
+      detectOutputFormat(ctx.settings.output),
+    );
+  },
+});
+
+export const managedAgentProjectVersionList = defineCommand({
+  description: { "en-US": "List directory project versions", "zh-CN": "列出目录项目版本" },
+  auth: "none",
+  usageArgs: "[--project <directory>] [--limit <n>] [--cursor <cursor>]",
+  flags: {
+    ...PROJECT_FLAG,
+    limit: {
+      type: "number",
+      valueHint: "<n>",
+      description: { "en-US": "Maximum versions to return", "zh-CN": "最多返回的版本数" },
+    },
+    cursor: {
+      type: "string",
+      valueHint: "<cursor>",
+      description: { "en-US": "Pagination cursor", "zh-CN": "分页游标" },
+    },
+  },
+  exampleArgs: ["", "--limit 20 --output json"],
+  async run(ctx) {
+    emitResult(
+      await createDirectoryWorkspaceVersionService(ctx.flags.project ?? ".").listVersions({
+        limit: ctx.flags.limit,
+        cursor: ctx.flags.cursor,
+      }),
+      detectOutputFormat(ctx.settings.output),
+    );
+  },
+});
+
+const VERSION_ID_FLAG = {
+  versionId: {
+    type: "string",
+    valueHint: "<full-version>",
+    required: true,
+    description: { "en-US": "Full project version ID", "zh-CN": "完整的项目版本 ID" },
+  },
+} satisfies FlagsDef;
+
+export const managedAgentProjectVersionPreview = defineCommand({
+  description: { "en-US": "Preview a directory project version", "zh-CN": "预览目录项目历史版本" },
+  auth: "none",
+  usageArgs: "--version-id <full-version> [--project <directory>]",
+  flags: { ...PROJECT_FLAG, ...VERSION_ID_FLAG },
+  exampleArgs: ["--version-id <full-version>"],
+  async run(ctx) {
+    emitResult(
+      await createDirectoryWorkspaceVersionService(ctx.flags.project ?? ".").previewVersion(
+        ctx.flags.versionId,
+      ),
+      detectOutputFormat(ctx.settings.output),
+    );
+  },
+});
+
+export const managedAgentProjectVersionRestore = defineCommand({
+  description: {
+    "en-US": "Restore a version to the project working directory",
+    "zh-CN": "将历史版本恢复到项目工作目录",
+  },
+  auth: "none",
+  risk: {
+    level: "high",
+    message: {
+      "en-US":
+        "This restores the full directory source to the working tree. Version history and remote State will not move.",
+      "zh-CN": "该操作会将完整目录源文件恢复到工作区；版本历史和远端 State 不会移动。",
+    },
+  },
+  usageArgs: "--version-id <full-version> [--project <directory>]",
+  flags: { ...PROJECT_FLAG, ...VERSION_ID_FLAG },
+  exampleArgs: ["--version-id <full-version>", "--version-id <full-version> --yes"],
+  async run(ctx) {
+    const service = createDirectoryWorkspaceVersionService(ctx.flags.project ?? ".");
+    const preview = await service.previewVersion(ctx.flags.versionId);
+    if (!preview.can_restore)
+      throw new BailianError(
+        preview.blockers[0] ?? preview.diagnostics[0]?.message ?? "Version cannot be restored.",
+        ExitCode.GENERAL,
+      );
+    if (ctx.settings.dryRun) {
+      emitResult(
+        { would_restore: ctx.flags.versionId, preview },
+        detectOutputFormat(ctx.settings.output),
+      );
+      return;
+    }
+    const restored = await service.restoreVersion(ctx.flags.versionId, {
+      headVersion: preview.base_head_version,
+      projectRevision: preview.base_project_revision,
+    });
+    emitResult(restored, detectOutputFormat(ctx.settings.output));
+  },
+});
+
+function safeInspection(result: {
+  project_root: string;
+  project_revision: string;
+  source_manifest_hash: string;
+  yaml_hash: string;
+  diagnostics: unknown[];
+  warnings: unknown[];
+  organization_moves: unknown[];
+  canonical_yaml: string;
+  before_yaml?: string;
+  can_build?: boolean;
+}) {
+  return {
+    project_root: result.project_root,
+    project_revision: result.project_revision,
+    source_manifest_hash: result.source_manifest_hash,
+    yaml_hash: result.yaml_hash,
+    diagnostics: result.diagnostics,
+    warnings: result.warnings,
+    organization_moves: result.organization_moves,
+    before_yaml: "before_yaml" in result ? result.before_yaml : undefined,
+    after_yaml: result.canonical_yaml,
+    can_build: "can_build" in result ? result.can_build : undefined,
+  };
+}

packages/commands/src/commands/managed-agent/workbench.ts

@@ -0,0 +1,90 @@
+import { defineCommand, detectOutputFormat, type FlagsDef } from "bailian-cli-core";
+import { emitResult } from "bailian-cli-runtime";
+import { CREDENTIALS_NOTE } from "./_engine/config-loader.ts";
+import { launchManagedAgentPlayground } from "./_engine/playground-launcher.ts";
+
+const PLAYGROUND_BASE_FLAGS = {
+  file: {
+    type: "string",
+    valueHint: "<path>",
+    description: {
+      "en-US": "Config file path (default: agents.yaml)",
+      "zh-CN": "配置文件路径（默认：agents.yaml）",
+    },
+  },
+  port: {
+    type: "number",
+    valueHint: "<n>",
+    description: {
+      "en-US": "Local port (default: 4848)",
+      "zh-CN": "本地端口（默认：4848）",
+    },
+  },
+  noOpen: {
+    type: "switch",
+    description: {
+      "en-US": "Do not open a browser automatically",
+      "zh-CN": "不自动打开浏览器",
+    },
+  },
+} satisfies FlagsDef;
+
+const PLAYGROUND_FLAGS = {
+  ...PLAYGROUND_BASE_FLAGS,
+  agent: {
+    type: "string",
+    valueHint: "<id>",
+    description: {
+      "en-US": "Agent to preview (required when the project declares multiple Agents)",
+      "zh-CN": "要预览的 Agent（项目包含多个 Agent 时需要指定）",
+    },
+  },
+} satisfies FlagsDef;
+
+const PLAYGROUND_NOTES = [
+  ...CREDENTIALS_NOTE,
+  {
+    "en-US":
+      "Session Preview requires Node.js 22+ and keeps using an agents.yaml source. Directory Workbench is available under managed-agent project workbench.",
+    "zh-CN":
+      "会话预览需要 Node.js 22+，并继续使用 agents.yaml；目录 Workbench 位于 managed-agent project workbench。",
+  },
+];
+
+export const managedAgentPlayground = defineCommand({
+  description: {
+    "en-US": "Launch a Session Preview for an agents.yaml Agent",
+    "zh-CN": "为 agents.yaml 中的 Agent 启动会话预览",
+  },
+  auth: "apiKey",
+  usageArgs: "[--file <path>] [--agent <id>] [--port <n>] [--no-open]",
+  flags: PLAYGROUND_FLAGS,
+  exampleArgs: ["", "--agent assistant", "--file agents.yaml --no-open"],
+  notes: PLAYGROUND_NOTES,
+  async run(ctx) {
+    const file = ctx.flags.file ?? "agents.yaml";
+    const port = ctx.flags.port ?? 4848;
+    if (ctx.settings.dryRun) {
+      emitResult(
+        {
+          would_launch: "playground",
+          config_file: file,
+          agent: ctx.flags.agent,
+          port,
+          open_browser: !ctx.flags.noOpen,
+        },
+        detectOutputFormat(ctx.settings.output),
+      );
+      return;
+    }
+    await launchManagedAgentPlayground({
+      file,
+      agent: ctx.flags.agent,
+      port,
+      open: !ctx.flags.noOpen,
+      surface: "preview",
+      client: ctx.client,
+      settings: ctx.settings,
+    });
+  },
+});

packages/commands/src/commands/speech/recognize.ts

@@ -14,9 +14,11 @@ import {
   speechRecognizePath,
   resolveAsrApi,
   buildAsrFlashRequest,
+  buildAsrContextMessages,
   buildAsyncAsrLanguageFields,
   collectAsrTranscriptionItems,
   extractAsrFlashText,
+  parseInstantVocabulary,
   type AsrApiRoute,
   type AsrFlashFamily,
   type OutputFormat,
@@ -73,8 +75,30 @@ const RECOGNIZE_FLAGS = {
     type: "string",
     valueHint: "<id>",
     description: {
-      "en-US": "Hot-word vocabulary ID for improved accuracy",
-      "zh-CN": "用于提升识别准确率的热词表 ID",
+      "en-US":
+        "Pre-built hot-word vocabulary ID (create it via `speech vocabulary create`). Its target_model must exactly match --model, otherwise it is silently ignored. Wider model support than --vocabulary, including Fun-ASR and Paraformer",
+      "zh-CN":
+        "预编译热词列表 ID（可用 `speech vocabulary create` 创建）。其 target_model 必须与 --model 完全一致，否则静默失效且无报错。支持模型比 --vocabulary 更广，含 Fun-ASR 与 Paraformer 系列",
+    },
+  },
+  vocabulary: {
+    type: "string",
+    valueHint: "<json>",
+    description: {
+      "en-US":
+        "Instant hot words as JSON object of word→weight, e.g. '{\"Fendouzhe\":4}'. Weight 1-5 (4 recommended; higher values can hurt other words), 50 for super hot word. No pre-built vocabulary needed. Takes effect only on Qwen-Audio-3.0-ASR-Flash models",
+      "zh-CN":
+        "即时热词，JSON 对象「热词→权重」，例如 '{\"奋斗者\":4}'。权重 1-5（推荐 4，过高会拖累其他词），50 表示超级热词。无需预先创建热词表。仅 Qwen-Audio-3.0-ASR-Flash 系列模型生效",
+    },
+  },
+  context: {
+    type: "string",
+    valueHint: "<text>",
+    description: {
+      "en-US":
+        "Context enhancement word list to improve accuracy on proper nouns; must contain the target words themselves (a topic description alone has little effect); max 400 chars. Takes effect only on Qwen-Audio-3.0-ASR-Flash and Fun-ASR-Flash models",
+      "zh-CN":
+        "上下文增强词表，提升专有名词准确率；须包含待识别的原词本身（只写主题描述效果有限），最长 400 字符。仅 Qwen-Audio-3.0-ASR-Flash 系列与 Fun-ASR-Flash 模型生效",
     },
   },
   channelId: {
@@ -110,9 +134,11 @@ function assertSyncFlashFlagsAllowed(
   const unsupported: string[] = [];
   if (flags.diarization === true) unsupported.push("--diarization");
   if (flags.speakerCount !== undefined) unsupported.push("--speaker-count");
-  // qwen3 sync Flash does not use vocabulary_id; input-audio Flash (fun-asr-flash* / qwen-audio-*-asr-flash) does
-  if (flashFamily === "qwen3" && flags.vocabularyId !== undefined) {
-    unsupported.push("--vocabulary-id");
+  // qwen3 sync Flash has no place for vocabulary_id / vocabulary / context in its body shape
+  if (flashFamily === "qwen3") {
+    if (flags.vocabularyId !== undefined) unsupported.push("--vocabulary-id");
+    if (flags.vocabulary !== undefined) unsupported.push("--vocabulary");
+    if (flags.context !== undefined) unsupported.push("--context");
   }
   if (flags.channelId !== undefined) unsupported.push("--channel-id");
   if (flags.async === true) unsupported.push("--async");
@@ -121,12 +147,34 @@ function assertSyncFlashFlagsAllowed(
   if (unsupported.length > 0) {
     throw new BailianError(
       `Model "${model}" uses sync Flash ASR and does not support: ${unsupported.join(", ")}.\n` +
-        `Hint: Use an async filetrans model (e.g. fun-asr, qwen3-asr-flash-filetrans) for those flags.`,
+        syncFlashUnsupportedHint(unsupported),
       ExitCode.USAGE,
     );
   }
 }
 
+/** Pick a hint that matches the rejected flags (vocab/context vs diarization/async/…). */
+function syncFlashUnsupportedHint(unsupported: string[]): string {
+  const vocabularyRelated = new Set(["--vocabulary", "--vocabulary-id", "--context"]);
+  const hasVocabularyRelated = unsupported.some((flag) => vocabularyRelated.has(flag));
+  const hasOtherFlags = unsupported.some((flag) => !vocabularyRelated.has(flag));
+
+  if (hasVocabularyRelated && !hasOtherFlags) {
+    return (
+      "Hint: Use qwen-audio-3.0-asr-flash (or an async filetrans model such as " +
+      "qwen-audio-3.0-asr-flash-filetrans) for vocabulary/context flags."
+    );
+  }
+  if (hasVocabularyRelated && hasOtherFlags) {
+    return (
+      "Hint: For vocabulary/context flags use qwen-audio-3.0-asr-flash or " +
+      "qwen-audio-3.0-asr-flash-filetrans; for the other flags use an async filetrans model " +
+      "(e.g. fun-asr)."
+    );
+  }
+  return "Hint: Use an async filetrans model (e.g. fun-asr, qwen3-asr-flash-filetrans) for those flags.";
+}
+
 export default defineCommand({
   description: {
     "en-US": "Recognize speech from audio files (FunAudio-ASR / Qwen-ASR Flash)",
@@ -141,6 +189,8 @@ export default defineCommand({
     "--url https://example.com/meeting.wav --diarization --speaker-count 3",
     "--url https://example.com/audio.mp3 --language zh",
     "--url https://example.com/audio.mp3 --vocabulary-id vocab-abc123",
+    '--url https://example.com/audio.mp3 --model qwen-audio-3.0-asr-flash-filetrans --vocabulary \'{"奋斗者":4,"鲸落":4}\'',
+    '--url https://example.com/audio.mp3 --model qwen-audio-3.0-asr-flash-filetrans --context "奋斗者号 鲸落 深海勇士"',
     "--url https://example.com/audio.mp3 --out result.json",
     "--url https://example.com/audio.mp3 --async --quiet",
     "--url https://example.com/audio.mp3 --model qwen-audio-3.0-asr-flash --language en",
@@ -198,6 +248,9 @@ export default defineCommand({
 
     const format = detectOutputFormat(settings.output);
 
+    const vocabulary =
+      flags.vocabulary !== undefined ? parseInstantVocabulary(flags.vocabulary) : undefined;
+
     // Auto-upload local files in parallel
     const resolvedUrls = await Promise.all(rawUrls.map((url) => ctx.client.uploadFile(url, model)));
 
@@ -210,6 +263,7 @@ export default defineCommand({
         model,
         route,
         resolvedUrls[0]!,
+        vocabulary,
       );
       return;
     }
@@ -223,16 +277,19 @@ export default defineCommand({
 
     const body: DashScopeASRRequest = {
       model,
-      input:
-        route.asyncInputStyle === "file_url"
+      input: {
+        ...(route.asyncInputStyle === "file_url"
           ? { file_url: resolvedUrls[0]! }
-          : { file_urls: resolvedUrls },
+          : { file_urls: resolvedUrls }),
+        ...(flags.context !== undefined ? { context: buildAsrContextMessages(flags.context) } : {}),
+      },
       parameters: {
         channel_id: channelId !== undefined ? [channelId] : [0],
         ...languageFields,
         diarization_enabled: diarization ? true : undefined,
         speaker_count: speakerCount,
         vocabulary_id: vocabularyId,
+        vocabulary,
       },
     };
 
@@ -260,6 +317,7 @@ async function handleSyncFlashMode(
   model: string,
   route: AsrApiRoute,
   audioUrl: string,
+  vocabulary: Record<string, number> | undefined,
 ): Promise<void> {
   const flashFamily = route.flashFamily as AsrFlashFamily;
   const body = buildAsrFlashRequest({
@@ -267,6 +325,8 @@ async function handleSyncFlashMode(
     audioUrl,
     language: flags.language,
     vocabularyId: flags.vocabularyId,
+    vocabulary,
+    context: flags.context,
     flashFamily,
   });
 

packages/commands/src/commands/speech/vocabulary/create.ts

@@ -0,0 +1,117 @@
+import {
+  defineCommand,
+  detectOutputFormat,
+  speechVocabularyPath,
+  buildVocabularyRequest,
+  createVocabulary,
+  type FlagsDef,
+  type ParsedFlags,
+} from "bailian-cli-core";
+import { emitResult, emitBare } from "bailian-cli-runtime";
+import {
+  VOCABULARY_BODY_FLAGS,
+  VOCABULARY_LIMIT_NOTES,
+  validateVocabularySource,
+  readVocabularyEntries,
+} from "./shared.ts";
+
+const CREATE_FLAGS = {
+  model: {
+    type: "string",
+    valueHint: "<model>",
+    description: {
+      "en-US":
+        "ASR model this vocabulary is built for (required). Must exactly match the --model passed to `speech recognize` later, otherwise the vocabulary is silently ignored",
+      "zh-CN":
+        "该热词表服务的 ASR 模型（必填）。必须与后续 `speech recognize` 的 --model 完全一致，否则热词表静默失效",
+    },
+    required: true,
+  },
+  prefix: {
+    type: "string",
+    valueHint: "<prefix>",
+    description: {
+      "en-US":
+        "Custom vocabulary prefix (required). Digits and lowercase letters only, max 10 chars",
+      "zh-CN": "热词表自定义前缀（必填）。仅允许数字和小写字母，最长 10 个字符",
+    },
+    required: true,
+  },
+  ...VOCABULARY_BODY_FLAGS,
+} satisfies FlagsDef;
+type CreateFlags = ParsedFlags<typeof CREATE_FLAGS>;
+
+export default defineCommand({
+  description: {
+    "en-US": "Create a precompiled hot-word vocabulary for ASR",
+    "zh-CN": "创建用于语音识别的预编译热词表",
+  },
+  auth: "apiKey",
+  usageArgs: "--model <model> --prefix <prefix> (--words <json> | --words-file <path>) [flags]",
+  flags: CREATE_FLAGS,
+  notes: [
+    {
+      "en-US":
+        "The --model must exactly match the --model used later with `speech recognize --vocabulary-id`; a mismatch causes silent failure with no error.",
+      "zh-CN":
+        "--model 必须与后续 `speech recognize --vocabulary-id` 使用的 --model 完全一致；不一致时热词表会静默失效且无报错。",
+    },
+    ...VOCABULARY_LIMIT_NOTES,
+  ],
+  exampleArgs: [
+    {
+      "en-US": '--model fun-asr --prefix demo --words \'{"Fendouzhe":4,"Jingluo":4}\'',
+      "zh-CN": '--model fun-asr --prefix demo --words \'{"奋斗者":4,"鲸落":4}\'',
+    },
+    {
+      "en-US":
+        '--model paraformer-v2 --prefix demo --words \'[{"text":"Fendouzhe","weight":4,"lang":"zh"}]\'',
+      "zh-CN":
+        '--model paraformer-v2 --prefix demo --words \'[{"text":"奋斗者","weight":4,"lang":"zh"}]\'',
+    },
+    {
+      "en-US": "--model fun-asr --prefix demo --words '{\"Fendouzhe\":4}' --lang zh",
+      "zh-CN": "--model fun-asr --prefix demo --words '{\"奋斗者\":4}' --lang zh",
+    },
+    "--model fun-asr --prefix demo --words-file ./hotwords.json",
+    {
+      "en-US": "--model fun-asr --prefix demo --words '{\"Fendouzhe\":4}' --quiet",
+      "zh-CN": "--model fun-asr --prefix demo --words '{\"奋斗者\":4}' --quiet",
+    },
+  ],
+  validate: (flags: CreateFlags) => validateVocabularySource(flags),
+  async run(ctx) {
+    const { settings, flags } = ctx;
+    const vocabulary = readVocabularyEntries(flags);
+    const format = detectOutputFormat(settings.output);
+
+    const request = buildVocabularyRequest("create_vocabulary", {
+      target_model: flags.model,
+      prefix: flags.prefix,
+      vocabulary,
+    });
+
+    if (settings.dryRun) {
+      emitResult(
+        {
+          endpoint: ctx.client.url(speechVocabularyPath()),
+          request,
+        },
+        format,
+      );
+      return;
+    }
+
+    const response = await createVocabulary(ctx.client, {
+      targetModel: flags.model,
+      prefix: flags.prefix,
+      vocabulary,
+    });
+
+    if (settings.quiet || format === "text") {
+      emitBare(response.output?.vocabulary_id ?? "");
+    } else {
+      emitResult(response, format);
+    }
+  },
+});

packages/commands/src/commands/speech/vocabulary/delete.ts

@@ -0,0 +1,60 @@
+import {
+  defineCommand,
+  detectOutputFormat,
+  speechVocabularyPath,
+  buildVocabularyRequest,
+  deleteVocabulary,
+  type FlagsDef,
+  type ParsedFlags,
+} from "bailian-cli-core";
+import { emitResult, emitBare } from "bailian-cli-runtime";
+import { VOCABULARY_ID_FLAG } from "./shared.ts";
+
+const DELETE_FLAGS = {
+  ...VOCABULARY_ID_FLAG,
+} satisfies FlagsDef;
+type DeleteFlags = ParsedFlags<typeof DELETE_FLAGS>;
+
+export default defineCommand({
+  description: {
+    "en-US": "Delete a precompiled hot-word vocabulary",
+    "zh-CN": "删除预编译热词表",
+  },
+  auth: "apiKey",
+  risk: {
+    level: "high",
+    message: {
+      "en-US": "This permanently deletes the specified hot-word vocabulary and cannot be undone.",
+      "zh-CN": "该操作会永久删除指定的热词表，且无法撤销。",
+    },
+  },
+  usageArgs: "--id <id>",
+  flags: DELETE_FLAGS,
+  exampleArgs: ["--id vocab-demo-xxx --dry-run", "--id vocab-demo-xxx --yes"],
+  async run(ctx) {
+    const { settings, flags } = ctx;
+    const vocabularyId = (flags as DeleteFlags).id;
+    const format = detectOutputFormat(settings.output);
+
+    if (settings.dryRun) {
+      emitResult(
+        {
+          endpoint: ctx.client.url(speechVocabularyPath()),
+          request: buildVocabularyRequest("delete_vocabulary", {
+            vocabulary_id: vocabularyId,
+          }),
+        },
+        format,
+      );
+      return;
+    }
+
+    const response = await deleteVocabulary(ctx.client, vocabularyId);
+
+    if (settings.quiet || format === "text") {
+      emitBare(vocabularyId);
+    } else {
+      emitResult(response, format);
+    }
+  },
+});

packages/commands/src/commands/speech/vocabulary/get.ts

@@ -0,0 +1,79 @@
+import {
+  defineCommand,
+  detectOutputFormat,
+  speechVocabularyPath,
+  buildVocabularyRequest,
+  queryVocabulary,
+  type FlagsDef,
+  type ParsedFlags,
+} from "bailian-cli-core";
+import { emitResult, emitBare } from "bailian-cli-runtime";
+import { VOCABULARY_ID_FLAG } from "./shared.ts";
+
+const GET_FLAGS = {
+  ...VOCABULARY_ID_FLAG,
+} satisfies FlagsDef;
+type GetFlags = ParsedFlags<typeof GET_FLAGS>;
+
+export default defineCommand({
+  description: {
+    "en-US": "Get details of a precompiled hot-word vocabulary",
+    "zh-CN": "查看预编译热词表详情",
+  },
+  auth: "apiKey",
+  usageArgs: "--id <id>",
+  flags: GET_FLAGS,
+  notes: [
+    {
+      "en-US":
+        "Use this command to confirm target_model before calling `speech recognize --vocabulary-id`; a model mismatch causes silent failure.",
+      "zh-CN":
+        "调用 `speech recognize --vocabulary-id` 前请用本命令确认 target_model；模型不一致会导致静默失效。",
+    },
+  ],
+  exampleArgs: ["--id vocab-demo-xxx", "--id vocab-demo-xxx --quiet"],
+  async run(ctx) {
+    const { settings, flags } = ctx;
+    const vocabularyId = (flags as GetFlags).id;
+    const format = detectOutputFormat(settings.output);
+
+    if (settings.dryRun) {
+      emitResult(
+        {
+          endpoint: ctx.client.url(speechVocabularyPath()),
+          request: buildVocabularyRequest("query_vocabulary", {
+            vocabulary_id: vocabularyId,
+          }),
+        },
+        format,
+      );
+      return;
+    }
+
+    const response = await queryVocabulary(ctx.client, vocabularyId);
+    const output = response.output;
+
+    if (settings.quiet) {
+      emitBare(output?.target_model ?? "");
+      return;
+    }
+
+    if (format === "text") {
+      emitBare(`vocabulary_id: ${vocabularyId}`);
+      emitBare(`status: ${output?.status ?? ""}`);
+      emitBare(`target_model: ${output?.target_model ?? ""}`);
+      emitBare(`gmt_create: ${output?.gmt_create ?? ""}`);
+      emitBare(`gmt_modified: ${output?.gmt_modified ?? ""}`);
+      const entries = output?.vocabulary ?? [];
+      if (entries.length > 0) {
+        emitBare("vocabulary:");
+        for (const entry of entries) {
+          const langPart = entry.lang ? ` lang=${entry.lang}` : "";
+          emitBare(`  ${entry.text} weight=${entry.weight}${langPart}`);
+        }
+      }
+    } else {
+      emitResult(response, format);
+    }
+  },
+});

packages/commands/src/commands/speech/vocabulary/list.ts

@@ -0,0 +1,116 @@
+import {
+  defineCommand,
+  detectOutputFormat,
+  speechVocabularyPath,
+  buildVocabularyRequest,
+  listVocabularies,
+  type FlagsDef,
+  type ParsedFlags,
+} from "bailian-cli-core";
+import { emitResult, emitBare } from "bailian-cli-runtime";
+
+const LIST_FLAGS = {
+  prefix: {
+    type: "string",
+    valueHint: "<prefix>",
+    description: {
+      "en-US": "Filter by vocabulary prefix",
+      "zh-CN": "按热词表前缀过滤",
+    },
+  },
+  page: {
+    type: "number",
+    valueHint: "<n>",
+    description: {
+      "en-US": "Page number, 1-based (default: 1). Mapped to API page_index (0-based) as page - 1",
+      "zh-CN": "页码，从 1 开始（默认：1）。映射为 API 的 page_index（从 0 开始）：page - 1",
+    },
+  },
+  pageSize: {
+    type: "number",
+    valueHint: "<n>",
+    description: {
+      "en-US": "Results per page (default: 10)",
+      "zh-CN": "每页结果数（默认：10）",
+    },
+  },
+} satisfies FlagsDef;
+type ListFlags = ParsedFlags<typeof LIST_FLAGS>;
+
+export default defineCommand({
+  description: {
+    "en-US": "List precompiled hot-word vocabularies",
+    "zh-CN": "列出预编译热词表",
+  },
+  auth: "apiKey",
+  usageArgs: "[--prefix <prefix>] [--page <n>] [--page-size <n>]",
+  flags: LIST_FLAGS,
+  notes: [
+    {
+      "en-US":
+        "List responses do not include target_model; use `speech vocabulary get` to inspect the model a vocabulary was built for.",
+      "zh-CN": "list 响应不含 target_model；要对齐模型请使用 `speech vocabulary get`。",
+    },
+    {
+      "en-US": "Vocabularies with status UNDEPLOYED are silently ignored by ASR.",
+      "zh-CN": "status 为 UNDEPLOYED 的热词表会被 ASR 静默忽略。",
+    },
+  ],
+  exampleArgs: ["", "--prefix demo", "--page 2 --page-size 20"],
+  validate: (flags: ListFlags) => {
+    if (flags.page !== undefined && flags.page < 1) {
+      return "--page must be >= 1.";
+    }
+    return undefined;
+  },
+  async run(ctx) {
+    const { settings, flags } = ctx;
+    const format = detectOutputFormat(settings.output);
+
+    const pageIndex = flags.page !== undefined ? flags.page - 1 : undefined;
+    const input: Record<string, unknown> = {};
+    if (flags.prefix !== undefined) input.prefix = flags.prefix;
+    if (pageIndex !== undefined) input.page_index = pageIndex;
+    if (flags.pageSize !== undefined) input.page_size = flags.pageSize;
+
+    if (settings.dryRun) {
+      emitResult(
+        {
+          endpoint: ctx.client.url(speechVocabularyPath()),
+          request: buildVocabularyRequest("list_vocabulary", input),
+        },
+        format,
+      );
+      return;
+    }
+
+    const response = await listVocabularies(ctx.client, {
+      prefix: flags.prefix,
+      pageIndex,
+      pageSize: flags.pageSize,
+    });
+
+    if (settings.quiet || format === "text") {
+      const items = response.output?.vocabulary_list ?? [];
+      if (items.length === 0) {
+        emitBare("No vocabularies found.");
+      } else {
+        let hasUndeployed = false;
+        for (const item of items) {
+          const id = item.vocabulary_id ?? "";
+          const status = item.status ?? "";
+          const modified = item.gmt_modified ?? "";
+          if (status === "UNDEPLOYED") hasUndeployed = true;
+          emitBare(`[${id}] ${status} ${modified}`.trimEnd());
+        }
+        if (hasUndeployed) {
+          emitBare(
+            "Note: UNDEPLOYED vocabularies are silently ignored by ASR. Use `speech vocabulary get` to inspect them.",
+          );
+        }
+      }
+    } else {
+      emitResult(response, format);
+    }
+  },
+});

packages/commands/src/commands/speech/vocabulary/shared.ts

@@ -0,0 +1,103 @@
+import {
+  readTextFromPathOrStdin,
+  parseVocabularyEntries,
+  UsageError,
+  type FlagsDef,
+  type ParsedFlags,
+  type VocabularyEntry,
+} from "bailian-cli-core";
+
+/** Shared --id flag for get / update / delete. */
+export const VOCABULARY_ID_FLAG = {
+  id: {
+    type: "string",
+    valueHint: "<id>",
+    description: {
+      "en-US": "Hot-word vocabulary ID (required)",
+      "zh-CN": "热词表 ID（必填）",
+    },
+    required: true,
+  },
+} satisfies FlagsDef;
+
+/** Shared hot-word body flags for create / update. */
+export const VOCABULARY_BODY_FLAGS = {
+  words: {
+    type: "string",
+    valueHint: "<json>",
+    description: {
+      "en-US":
+        "Hot words as JSON object of word→weight, e.g. '{\"Fendouzhe\":4}'; or the API entry array for per-entry lang. Weight 1-5 (4 recommended); when the vocabulary target_model is a Qwen-Audio-3.0-ASR-Flash series model, 50 is also allowed as super hot word. Or use --words-file",
+      "zh-CN":
+        "热词，JSON 对象「热词→权重」，例如 '{\"奋斗者\":4}'；需要逐条指定语言时可传 API 的条目数组。权重 1-5（推荐 4）；当热词表的 target_model 为 Qwen-Audio-3.0-ASR-Flash 系列时还可使用 50（超级热词）。也可使用 --words-file",
+    },
+  },
+  wordsFile: {
+    type: "string",
+    valueHint: "<path>",
+    description: {
+      "en-US": "JSON file with the hot words (use - for stdin)",
+      "zh-CN": "包含热词的 JSON 文件（使用 - 从 stdin 读取）",
+    },
+  },
+  lang: {
+    type: "string",
+    valueHint: "<code>",
+    description: {
+      "en-US":
+        "Language code applied to every hot word when using object form (optional; ignored for array form). Paraformer: zh/en/ja/yue/ko/de/fr/ru; Fun-ASR: zh/en/ja",
+      "zh-CN":
+        "对象形态时应用到所有热词的语言代码（选填；数组形态忽略）。Paraformer 支持 zh/en/ja/yue/ko/de/fr/ru；Fun-ASR 支持 zh/en/ja",
+    },
+  },
+} satisfies FlagsDef;
+
+type VocabularySourceFlags = ParsedFlags<typeof VOCABULARY_BODY_FLAGS>;
+
+/** Cross-flag validation for --words / --words-file. */
+export function validateVocabularySource(flags: VocabularySourceFlags): string | undefined {
+  if (!flags.words && !flags.wordsFile) {
+    return "Provide --words or --words-file.";
+  }
+  if (flags.words && flags.wordsFile) {
+    return "Use either --words or --words-file, not both.";
+  }
+  if (!flags.words) {
+    return undefined;
+  }
+  try {
+    parseVocabularyEntries(flags.words, flags.lang);
+  } catch (error) {
+    if (error instanceof UsageError) {
+      return error.message;
+    }
+    throw error;
+  }
+  return undefined;
+}
+
+/** Read and parse vocabulary entries from flag or file. */
+export function readVocabularyEntries(flags: VocabularySourceFlags): VocabularyEntry[] {
+  const raw = flags.wordsFile ? readTextFromPathOrStdin(flags.wordsFile) : (flags.words as string);
+  return parseVocabularyEntries(raw, flags.lang);
+}
+
+/** Shared notes covering account limits and silent-failure pitfalls. */
+export const VOCABULARY_LIMIT_NOTES = [
+  {
+    "en-US":
+      "Each account may have at most 10 vocabularies; updates must be at least 5 minutes apart. See improve-asr-accuracy for full limits.",
+    "zh-CN": "每个账号最多 10 个热词表；两次更新间隔至少 5 分钟。完整限制见 improve-asr-accuracy。",
+  },
+  {
+    "en-US":
+      "Hot-word vocabularies are not supported in Singapore sub-workspaces; the server error is passed through as-is.",
+    "zh-CN": "新加坡子业务空间不支持热词表；服务端错误会原样透传。",
+  },
+  {
+    "en-US":
+      "Weight 1-5 (4 recommended); when the vocabulary target_model is a Qwen-Audio-3.0-ASR-Flash series model, 50 is also allowed as super hot word.",
+    "zh-CN":
+      "权重 1-5（推荐 4）；当热词表的 target_model 为 Qwen-Audio-3.0-ASR-Flash 系列时还可使用 50（超级热词）。",
+  },
+] as const;

packages/commands/src/commands/speech/vocabulary/update.ts

@@ -0,0 +1,90 @@
+import {
+  defineCommand,
+  detectOutputFormat,
+  speechVocabularyPath,
+  buildVocabularyRequest,
+  updateVocabulary,
+  type FlagsDef,
+  type ParsedFlags,
+} from "bailian-cli-core";
+import { emitResult, emitBare } from "bailian-cli-runtime";
+import {
+  VOCABULARY_ID_FLAG,
+  VOCABULARY_BODY_FLAGS,
+  VOCABULARY_LIMIT_NOTES,
+  validateVocabularySource,
+  readVocabularyEntries,
+} from "./shared.ts";
+
+const UPDATE_FLAGS = {
+  ...VOCABULARY_ID_FLAG,
+  ...VOCABULARY_BODY_FLAGS,
+} satisfies FlagsDef;
+type UpdateFlags = ParsedFlags<typeof UPDATE_FLAGS>;
+
+export default defineCommand({
+  description: {
+    "en-US": "Replace the contents of a precompiled hot-word vocabulary",
+    "zh-CN": "完全替换预编译热词表的内容",
+  },
+  auth: "apiKey",
+  risk: {
+    level: "high",
+    message: {
+      "en-US":
+        "This fully replaces all hot words in the vocabulary. Entries not listed will be discarded and cannot be undone.",
+      "zh-CN": "该操作会完全替换热词表中的全部词条。未列出的词将被丢弃，且无法撤销。",
+    },
+  },
+  usageArgs: "--id <id> (--words <json> | --words-file <path>) [flags]",
+  flags: UPDATE_FLAGS,
+  notes: [
+    {
+      "en-US":
+        "update is a full replace, not an append. Prefer --dry-run first to preview the complete vocabulary that will be written.",
+      "zh-CN": "update 是完全替换，不是增量追加。建议先用 --dry-run 预览将要写入的完整词表。",
+    },
+    ...VOCABULARY_LIMIT_NOTES,
+  ],
+  exampleArgs: [
+    {
+      "en-US": "--id vocab-demo-xxx --words '{\"Fendouzhe\":4}' --dry-run",
+      "zh-CN": "--id vocab-demo-xxx --words '{\"奋斗者\":4}' --dry-run",
+    },
+    {
+      "en-US": '--id vocab-demo-xxx --words \'{"Fendouzhe":4,"Jingluo":4}\' --yes',
+      "zh-CN": '--id vocab-demo-xxx --words \'{"奋斗者":4,"鲸落":4}\' --yes',
+    },
+  ],
+  validate: (flags: UpdateFlags) => validateVocabularySource(flags),
+  async run(ctx) {
+    const { settings, flags } = ctx;
+    const vocabularyId = flags.id;
+    const vocabulary = readVocabularyEntries(flags);
+    const format = detectOutputFormat(settings.output);
+
+    const request = buildVocabularyRequest("update_vocabulary", {
+      vocabulary_id: vocabularyId,
+      vocabulary,
+    });
+
+    if (settings.dryRun) {
+      emitResult(
+        {
+          endpoint: ctx.client.url(speechVocabularyPath()),
+          request,
+        },
+        format,
+      );
+      return;
+    }
+
+    const response = await updateVocabulary(ctx.client, vocabularyId, vocabulary);
+
+    if (settings.quiet || format === "text") {
+      emitBare(vocabularyId);
+    } else {
+      emitResult(response, format);
+    }
+  },
+});

packages/commands/src/commands/video/edit.ts

@@ -197,7 +197,7 @@ export default defineCommand({
 
     // --- Build request body ---
     const promptExtend = resolveBooleanFlag(flags.promptExtend, undefined, "prompt-extend");
-    const watermark = resolveWatermark(flags.watermark);
+    const watermark = resolveWatermark(flags.watermark, settings.watermark);
 
     const body: DashScopeVideoEditRequest = {
       model,

packages/commands/src/commands/video/generate.ts

@@ -208,7 +208,7 @@ export default defineCommand({
       resolvedFileUrl = await ctx.client.uploadFile(fileUrl, model);
     }
 
-    const watermark = resolveWatermark(flags.watermark);
+    const watermark = resolveWatermark(flags.watermark, settings.watermark);
     const promptExtend = resolveBooleanFlag(flags.promptExtend, undefined, "prompt-extend");
 
     const body: DashScopeVideoRequest = {

packages/commands/src/commands/video/ref.ts

@@ -249,7 +249,7 @@ export default defineCommand({
 
     // --- Build request body ---
     const promptExtend = resolveBooleanFlag(flags.promptExtend, undefined, "prompt-extend");
-    const watermark = resolveWatermark(flags.watermark);
+    const watermark = resolveWatermark(flags.watermark, settings.watermark);
 
     const body: DashScopeVideoRefRequest = {
       model,

packages/commands/src/index.ts

@@ -73,6 +73,11 @@ export { default as mcpTools } from "./commands/mcp/tools.ts";
 export { default as searchWeb } from "./commands/search/web.ts";
 export { default as speechSynthesize } from "./commands/speech/synthesize.ts";
 export { default as speechRecognize } from "./commands/speech/recognize.ts";
+export { default as speechVocabularyCreate } from "./commands/speech/vocabulary/create.ts";
+export { default as speechVocabularyList } from "./commands/speech/vocabulary/list.ts";
+export { default as speechVocabularyGet } from "./commands/speech/vocabulary/get.ts";
+export { default as speechVocabularyUpdate } from "./commands/speech/vocabulary/update.ts";
+export { default as speechVocabularyDelete } from "./commands/speech/vocabulary/delete.ts";
 export { default as fileUpload } from "./commands/file/upload.ts";
 export { default as consoleCall } from "./commands/console/call.ts";
 export { default as usageFree } from "./commands/usage/free.ts";
@@ -137,6 +142,20 @@ export { default as managedAgentValidate } from "./commands/managed-agent/valida
 export { default as managedAgentPlan } from "./commands/managed-agent/plan.ts";
 export { default as managedAgentApply } from "./commands/managed-agent/apply.ts";
 export { default as managedAgentDestroy } from "./commands/managed-agent/destroy.ts";
+export { managedAgentPlayground } from "./commands/managed-agent/workbench.ts";
+export {
+  managedAgentProjectBuild,
+  managedAgentProjectInit,
+  managedAgentProjectPublish,
+  managedAgentProjectValidate,
+  managedAgentProjectVersionDisable,
+  managedAgentProjectVersionEnable,
+  managedAgentProjectVersionList,
+  managedAgentProjectVersionPreview,
+  managedAgentProjectVersionRestore,
+  managedAgentProjectVersionStatus,
+  managedAgentProjectWorkbench,
+} from "./commands/managed-agent/project.ts";
 export { default as managedAgentStateList } from "./commands/managed-agent/state-list.ts";
 export { default as managedAgentStateShow } from "./commands/managed-agent/state-show.ts";
 export { default as managedAgentStateRm } from "./commands/managed-agent/state-rm.ts";

packages/commands/tests/config-shared.test.ts

@@ -33,3 +33,9 @@ test("default-speech-recognition-model alias accepts an ASR model ID", () => {
     "qwen-audio-3.0-asr-flash",
   );
 });
+
+test("watermark config accepts only boolean text and stores a boolean", () => {
+  expect(validateAndCoerce("watermark", "false")).toBe(false);
+  expect(validateAndCoerce("watermark", "TRUE")).toBe(true);
+  expect(() => validateAndCoerce("watermark", "yes")).toThrow(/true or false/i);
+});

packages/commands/tests/config-ui.test.ts

@@ -200,8 +200,10 @@ test("GET /api/config 返回全部 profile、明文密钥与持久化激活项",
     expect(res.json.keys).toContain("console_site");
     expect(res.json.keys).toContain("telemetry");
     expect(res.json.keys).toContain("default_speech_recognition_model");
+    expect(res.json.keys).toContain("watermark");
     expect(res.json.enums.console_site).toEqual(["domestic", "international"]);
     expect(res.json.booleanKeys).toContain("telemetry");
+    expect(res.json.booleanKeys).toContain("watermark");
     // Default field hints are surfaced as prefilled values in the UI.
     expect(res.json.fieldDefaults.default_image_model).toBe("qwen-image-3.0");
     expect(res.json.fieldDefaults.default_text_model).toBe("qwen3.8-max");

packages/commands/tests/credentials-bridge.test.ts

@@ -104,6 +104,20 @@ test("inject:base_url 已带后缀不重复拼;非空字面量 base_url 保留",
   expect(literal.bailian.base_url).toBe("https://custom.example.com/api/v1/agentstudio");
 });
 
+test("inject:目录项目可强制使用 Bailian CLI 解析出的 base_url", () => {
+  const providers = {
+    bailian: {
+      api_key: "from-project-env",
+      base_url: "https://project-env.example.com/api/v1/agentstudio",
+    },
+  };
+  injectProviderCredentials(providers, makeHost({ apiCred: bailianCred() }), {
+    overrideBaseUrl: true,
+  });
+  expect(providers.bailian.api_key).toBe("sk-auth-chain");
+  expect(providers.bailian.base_url).toBe("https://dashscope.aliyuncs.com/api/v1/agentstudio");
+});
+
 test("inject:base_url 尾斜杠被规范化,不产生双斜杠", () => {
   const providers = { bailian: { api_key: "", base_url: "" } };
   injectProviderCredentials(

packages/commands/tests/e2e/config.e2e.test.ts

@@ -64,10 +64,53 @@ describe("e2e: config", () => {
       config_file?: string;
       base_url?: string;
       timeout?: number;
+      watermark?: boolean;
     }>(stdout);
     expect(data.config_file).toBeDefined();
     expect(data.base_url).toBeDefined();
     expect(data.timeout).toBeDefined();
+    expect(data.watermark).toBe(true);
+  });
+
+  test("config set 将 watermark 作为 boolean 写入并由 config show 读回", async () => {
+    const configDir = mkdtempSync(join(tmpdir(), "bl-config-watermark-"));
+    try {
+      const env = { BAILIAN_CONFIG_DIR: configDir };
+      const setResult = await runCommandE2e(
+        CONFIG_ROUTES,
+        [
+          "config",
+          "set",
+          "--config",
+          "media",
+          "--key",
+          "watermark",
+          "--value",
+          "false",
+          "--output",
+          "json",
+        ],
+        env,
+      );
+      expect(setResult.exitCode, setResult.stderr).toBe(0);
+      expect(parseStdoutJson<{ watermark?: boolean }>(setResult.stdout).watermark).toBe(false);
+
+      const persisted = JSON.parse(readFileSync(join(configDir, "config.json"), "utf8")) as Record<
+        string,
+        Record<string, unknown>
+      >;
+      expect(persisted.media?.watermark).toBe(false);
+
+      const showResult = await runCommandE2e(
+        CONFIG_ROUTES,
+        ["config", "show", "--config", "media", "--output", "json"],
+        env,
+      );
+      expect(showResult.exitCode, showResult.stderr).toBe(0);
+      expect(parseStdoutJson<{ watermark?: boolean }>(showResult.stdout).watermark).toBe(false);
+    } finally {
+      rmSync(configDir, { recursive: true, force: true });
+    }
   });
 
   test("config show --output text", async () => {

packages/commands/tests/e2e/managed-agent-init.e2e.test.ts

@@ -0,0 +1,112 @@
+import { mkdir, mkdtemp, readFile, realpath, rm, stat, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { dirname, join } from "node:path";
+import { fileURLToPath } from "node:url";
+import { runNodeMain } from "e2e/runner";
+import { afterEach, describe, expect, test } from "vite-plus/test";
+import { parseStdoutJson } from "./helpers.ts";
+
+const directories: string[] = [];
+
+async function temporaryDirectory() {
+  const directory = await realpath(await mkdtemp(join(tmpdir(), "bailian-yaml-init-")));
+  directories.push(directory);
+  return directory;
+}
+
+afterEach(async () => {
+  for (const directory of directories.splice(0)) {
+    await rm(directory, { recursive: true, force: true });
+  }
+});
+
+async function runInit(directory: string, args: string[] = []) {
+  return runNodeMain(
+    fileURLToPath(new URL("./harness/main.ts", import.meta.url)),
+    ["managed-agent", "init", ...args],
+    {
+      cwd: directory,
+      env: {
+        BAILIAN_CONFIG_DIR: await temporaryDirectory(),
+        BAILIAN_E2E_ROUTES: JSON.stringify([
+          { path: "managed-agent init", export: "managedAgentInit" },
+        ]),
+      },
+    },
+  );
+}
+
+describe("e2e: managed-agent init output path", () => {
+  test.each(["default", "relative", "absolute"] as const)(
+    "reports the absolute YAML path for a %s output path",
+    async (pathKind) => {
+      const directory = await temporaryDirectory();
+      const relativePath =
+        pathKind === "default" ? "agents.yaml" : join("config files", "custom agents.yaml");
+      const outputPath = join(directory, relativePath);
+      await mkdir(dirname(outputPath), { recursive: true });
+      const args =
+        pathKind === "default"
+          ? []
+          : ["--file", pathKind === "absolute" ? outputPath : relativePath];
+
+      const result = await runInit(directory, args);
+
+      expect(result.exitCode, result.stderr).toBe(0);
+      expect(result.stdout.split(/\r?\n/)[0]).toBe(`Created ${outputPath}`);
+      expect(await readFile(outputPath, "utf8")).toContain("agents:");
+    },
+  );
+
+  test("preserves the JSON output contract", async () => {
+    const directory = await temporaryDirectory();
+    const result = await runInit(directory, ["--file", "custom.yaml", "--output", "json"]);
+
+    expect(result.exitCode, result.stderr).toBe(0);
+    expect(parseStdoutJson(result.stdout)).toEqual({
+      created: "custom.yaml",
+      provider: "bailian",
+      agent: "assistant",
+    });
+    expect((await stat(join(directory, "custom.yaml"))).isFile()).toBe(true);
+  });
+
+  test("keeps dry-run read-only without reporting a created file", async () => {
+    const directory = await temporaryDirectory();
+    const result = await runInit(directory, ["--dry-run", "--output", "json"]);
+
+    expect(result.exitCode, result.stderr).toBe(0);
+    expect(parseStdoutJson(result.stdout)).toEqual({
+      would_create: "agents.yaml",
+      provider: "bailian",
+      agent: "assistant",
+      would_update_gitignore: true,
+    });
+    expect(await stat(join(directory, "agents.yaml")).catch(() => null)).toBeNull();
+    expect(await stat(join(directory, ".gitignore")).catch(() => null)).toBeNull();
+  });
+
+  test.each(["default", "relative", "absolute"] as const)(
+    "reports the absolute existing %s path without overwriting the YAML",
+    async (pathKind) => {
+      const directory = await temporaryDirectory();
+      const relativePath =
+        pathKind === "default" ? "agents.yaml" : join("config files", "custom agents.yaml");
+      const outputPath = join(directory, relativePath);
+      await mkdir(dirname(outputPath), { recursive: true });
+      await writeFile(outputPath, "Keep user configuration.\n");
+      const args =
+        pathKind === "default"
+          ? []
+          : ["--file", pathKind === "absolute" ? outputPath : relativePath];
+
+      const result = await runInit(directory, args);
+
+      expect(result.exitCode).toBe(2);
+      expect(result.stderr).toContain(`${outputPath} already exists.`);
+      expect(result.stderr).toContain("Pass --force to overwrite.");
+      expect(result.stdout).not.toContain("Created ");
+      expect(await readFile(outputPath, "utf8")).toBe("Keep user configuration.\n");
+    },
+  );
+});

packages/commands/tests/e2e/managed-agent-project-init.e2e.test.ts

@@ -0,0 +1,247 @@
+import {
+  cp,
+  mkdir,
+  mkdtemp,
+  readFile,
+  realpath,
+  rm,
+  stat,
+  symlink,
+  writeFile,
+} from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { fileURLToPath } from "node:url";
+import { runNodeMain } from "e2e/runner";
+import { afterEach, describe, expect, test } from "vite-plus/test";
+import { parseStdoutJson, runCommandHelp } from "./helpers.ts";
+
+const routes = {
+  "managed-agent project init": "managedAgentProjectInit",
+  "managed-agent project build": "managedAgentProjectBuild",
+  "managed-agent project publish": "managedAgentProjectPublish",
+  "managed-agent project validate": "managedAgentProjectValidate",
+} as const;
+const directories: string[] = [];
+
+async function temporaryDirectory() {
+  const directory = await realpath(await mkdtemp(join(tmpdir(), "bailian-project-init-")));
+  directories.push(directory);
+  return directory;
+}
+
+afterEach(async () => {
+  for (const directory of directories.splice(0)) {
+    await rm(directory, { recursive: true, force: true });
+  }
+});
+
+function runInit(directory: string, args: string[] = []) {
+  return runProject(directory, "init", args);
+}
+
+async function runProject(directory: string, subcommand: string, args: string[] = [], json = true) {
+  const configRoot = await temporaryDirectory();
+  return runNodeMain(
+    fileURLToPath(new URL("./harness/main.ts", import.meta.url)),
+    ["managed-agent", "project", subcommand, ...args, ...(json ? ["--output", "json"] : [])],
+    {
+      cwd: directory,
+      env: {
+        BAILIAN_CONFIG_DIR: configRoot,
+        BAILIAN_E2E_ROUTES: JSON.stringify(
+          Object.entries(routes).map(([path, exportName]) => ({ path, export: exportName })),
+        ),
+      },
+    },
+  );
+}
+
+describe("e2e: managed-agent project init directory defaults", () => {
+  test("build links copied resources and reports ambiguous environment bindings without writing", async () => {
+    const directory = await temporaryDirectory();
+    const initialized = await runInit(directory);
+    expect(initialized.exitCode, initialized.stderr).toBe(0);
+    const root = join(directory, "managed-agent");
+    for (const [section, id] of [
+      ["skills", "example-skill"],
+      ["files", "example-file"],
+      ["environments", "example-env"],
+      ["vaults", "example-vault"],
+    ] as const) {
+      await cp(
+        join(root, "agents/assistant", section, "_examples", id),
+        join(root, "agents/assistant", section, id),
+        { recursive: true },
+      );
+    }
+    const agentPath = join(root, "agents/assistant/agent.json");
+    const original = await readFile(agentPath, "utf8");
+    const preview = await runProject(root, "build", ["--dry-run"]);
+    expect(preview.exitCode, preview.stderr).toBe(0);
+    expect(parseStdoutJson<{ can_build: boolean }>(preview.stdout).can_build).toBe(true);
+    expect(await readFile(agentPath, "utf8")).toBe(original);
+    const built = await runProject(root, "build");
+    expect(built.exitCode, built.stderr).toBe(0);
+    const agent = JSON.parse(await readFile(agentPath, "utf8"));
+    expect(agent).toMatchObject({
+      environment: "example-env",
+      vault: "example-vault",
+      skills: ["example-skill"],
+      files: [{ file: "example-file", mount_path: "/mnt/example.md" }],
+    });
+    const alternatePath = join(root, "agents/assistant/environments/alternate");
+    await mkdir(alternatePath);
+    await writeFile(
+      join(alternatePath, "environment.json"),
+      JSON.stringify({ id: "alternate", config: { type: "cloud" } }),
+    );
+    delete agent.environment;
+    await writeFile(agentPath, JSON.stringify(agent));
+    const beforeConflict = await readFile(agentPath, "utf8");
+    const buildPath = join(root, ".openagentpack/build/agents.yaml");
+    const beforeBuild = await readFile(buildPath, "utf8");
+    for (const json of [false, true]) {
+      const conflict = await runProject(root, "build", [], json);
+      expect(conflict.exitCode).toBe(1);
+      expect(conflict.stderr).toContain("multiple local environment resources");
+      expect(conflict.stderr).toContain("Set 'environment' explicitly");
+      expect(conflict.stderr).not.toMatch(/\p{Script=Han}/u);
+    }
+    expect(await readFile(agentPath, "utf8")).toBe(beforeConflict);
+    expect(await readFile(buildPath, "utf8")).toBe(beforeBuild);
+  });
+
+  test("build checks directories and writes without confirmation while publish stays gated", async () => {
+    const directory = await temporaryDirectory();
+    const initialized = await runInit(directory);
+    expect(initialized.exitCode, initialized.stderr).toBe(0);
+    const root = join(directory, "managed-agent");
+    const nested = join(root, "agents/assistant/skills");
+    for (const args of [[], ["--dry-run"]]) {
+      const result = await runProject(nested, "build", args, false);
+      expect(result.exitCode, result.stderr).toBe(2);
+      expect(result.stderr).toContain("Not a project root:");
+      expect(result.stderr).not.toMatch(/\p{Script=Han}/u);
+      expect(result.stderr).toContain(`cd '${root}'`);
+      expect(result.stderr).not.toContain("high-risk");
+      expect(result.stderr).not.toContain("Usage:");
+    }
+    const explicit = await runProject(root, "build", ["--project", nested]);
+    expect(explicit.exitCode).toBe(2);
+    expect(explicit.stderr).toContain(`--project '${root}'`);
+    const noMarker = await runProject(directory, "build", [], false);
+    expect(noMarker.exitCode).toBe(2);
+    expect(noMarker.stderr).toContain("project.json");
+    expect(noMarker.stderr).not.toMatch(/\p{Script=Han}/u);
+    expect(noMarker.stderr).not.toContain("high-risk");
+    expect(await stat(join(nested, ".openagentpack")).catch(() => null)).toBeNull();
+
+    const preview = await runProject(root, "build", ["--dry-run"]);
+    expect(preview.exitCode, preview.stderr).toBe(0);
+    expect(await stat(join(root, ".openagentpack/build")).catch(() => null)).toBeNull();
+    const built = await runProject(root, "build");
+    expect(built.exitCode, built.stderr).toBe(0);
+    expect(built.stderr).not.toContain("requires_confirmation");
+    expect((await stat(join(root, ".openagentpack/build/agents.yaml"))).isFile()).toBe(true);
+    const explicitValid = await runProject(nested, "build", ["--project", root]);
+    expect(explicitValid.exitCode, explicitValid.stderr).toBe(0);
+    const storePath = join(root, ".openagentpack/versions/project/store.json");
+    const beforePublish = await readFile(storePath, "utf8");
+    const publish = await runProject(root, "publish");
+    expect(publish.exitCode).toBe(7);
+    expect(publish.stderr).toContain("requires_confirmation");
+    expect(await readFile(storePath, "utf8")).toBe(beforePublish);
+  });
+
+  test("only publish help includes confirmation; build keeps dry-run", async () => {
+    const build = await runCommandHelp(routes, ["managed-agent", "project", "build", "--help"]);
+    expect(build.stderr).not.toContain("--yes");
+    expect(build.stderr).toContain("--dry-run");
+    const publish = await runCommandHelp(routes, ["managed-agent", "project", "publish", "--help"]);
+    expect(publish.stderr).toContain("--yes");
+    const directory = await temporaryDirectory();
+    const removedFlag = await runProject(directory, "build", ["--yes"]);
+    expect(removedFlag.exitCode).toBe(2);
+    expect(removedFlag.stderr).toMatch(/Unknown flag.*--yes/);
+  });
+
+  test("nested build and validate explain the project root without changing directories", async () => {
+    const directory = await temporaryDirectory();
+    const initialized = await runInit(directory);
+    expect(initialized.exitCode, initialized.stderr).toBe(0);
+    const root = join(directory, "managed-agent");
+    const nested = join(root, "agents/assistant/skills");
+    for (const command of ["build", "validate"]) {
+      const result = await runProject(nested, command, ["--dry-run"]);
+      expect(result.exitCode).not.toBe(0);
+      expect(result.stderr).toContain("Not a project root:");
+      expect(result.stderr).not.toMatch(/\p{Script=Han}/u);
+      expect(result.stderr).toContain(`cd '${root}'`);
+      expect(result.stderr).toContain(`--project '${root}'`);
+      expect(result.stderr).not.toContain("ERR_MODULE_NOT_FOUND");
+    }
+    expect(await stat(join(nested, ".openagentpack")).catch(() => null)).toBeNull();
+    const corrected = await runProject(nested, "build", ["--project", root, "--dry-run"]);
+    expect(corrected.exitCode, corrected.stderr).toBe(0);
+  });
+
+  test("help describes the subdirectory default and explicit in-place initialization", async () => {
+    const result = await runCommandHelp(routes, ["managed-agent", "project", "init", "--help"]);
+    expect(result.exitCode).toBe(0);
+    expect(result.stderr).toContain("./managed-agent");
+    expect(result.stderr).toContain("--project .");
+  });
+
+  test("default dry-run reports the child directory without creating it", async () => {
+    const directory = await temporaryDirectory();
+    const result = await runInit(directory, ["--dry-run"]);
+    expect(result.exitCode, result.stderr).toBe(0);
+    expect(parseStdoutJson(result.stdout)).toEqual({ would_initialize_project: "./managed-agent" });
+    expect(await stat(join(directory, "managed-agent")).catch(() => null)).toBeNull();
+  });
+
+  test("initializes only the child, ignores parent source, and rejects repeated init without overwriting", async () => {
+    const directory = await temporaryDirectory();
+    const parentYaml = "not a valid project declaration";
+    await writeFile(join(directory, "agents.yaml"), parentYaml);
+    if (process.platform !== "win32") {
+      await symlink("missing-instructions.md", join(directory, "CLAUDE.md"));
+    }
+    const result = await runInit(directory);
+    expect(result.exitCode, result.stderr).toBe(0);
+    const projectRoot = join(directory, "managed-agent");
+    const initialized = parseStdoutJson<{
+      project_root: string;
+      baseline_version: string;
+      converted_from_yaml: boolean;
+    }>(result.stdout);
+    expect(initialized.project_root).toBe(projectRoot);
+    expect(initialized.baseline_version).toHaveLength(64);
+    expect(initialized.converted_from_yaml).toBe(false);
+    expect(await stat(join(directory, "project.json")).catch(() => null)).toBeNull();
+    expect(await readFile(join(directory, "agents.yaml"), "utf8")).toBe(parentYaml);
+    expect((await stat(join(projectRoot, "agents/assistant/agent.json"))).isFile()).toBe(true);
+    expect(
+      (await stat(join(projectRoot, ".openagentpack/versions/project/store.json"))).isFile(),
+    ).toBe(true);
+    const instructions = join(projectRoot, "agents/assistant/instructions.md");
+    await writeFile(instructions, "user changes");
+    const repeated = await runInit(directory);
+    expect(repeated.exitCode).not.toBe(0);
+    expect(repeated.stderr).toContain("already exists");
+    expect(await readFile(instructions, "utf8")).toBe("user changes");
+  });
+
+  test("explicit paths, including dot, remain exact project roots", async () => {
+    for (const target of ["custom-agent", "."]) {
+      const directory = await temporaryDirectory();
+      const result = await runInit(directory, ["--project", target]);
+      expect(result.exitCode, result.stderr).toBe(0);
+      expect(parseStdoutJson<{ project_root: string }>(result.stdout).project_root).toBe(
+        join(directory, target),
+      );
+      expect(await stat(join(directory, "managed-agent")).catch(() => null)).toBeNull();
+    }
+  });
+});

packages/commands/tests/e2e/managed-agent.e2e.test.ts

@@ -1,4 +1,4 @@
-import { mkdir, mkdtemp, readFile, rm, writeFile } from "node:fs/promises";
+import { mkdir, mkdtemp, readFile, rm, stat, writeFile } from "node:fs/promises";
 import http from "node:http";
 import type { AddressInfo } from "node:net";
 import { tmpdir } from "node:os";
@@ -9,7 +9,7 @@ import {
   resolveProjectConfig,
   type ResourceAddress,
 } from "@openagentpack/sdk";
-import { describe, expect, test } from "vite-plus/test";
+import { afterEach, describe, expect, test } from "vite-plus/test";
 import { parse } from "yaml";
 import { e2eFixturesDir, parseStdoutJson, runCommandHelp, runCommandE2e } from "./helpers.ts";
 import { MANAGED_AGENT_ROUTES } from "./topic-routes.ts";
@@ -29,6 +29,13 @@ const DEPLOYMENT_SAFETY_DIAGNOSTIC_CODES = [
   "bailian.deployment.file.mount_path.required",
   "bailian.deployment.file.mount_path.duplicate",
 ];
+const projectDirectories: string[] = [];
+
+afterEach(async () => {
+  for (const directory of projectDirectories.splice(0)) {
+    await rm(directory, { recursive: true, force: true });
+  }
+});
 
 async function writeLocalSkillDirectory(parentDirectory: string): Promise<string> {
   const skillDirectory = join(parentDirectory, "local-report");
@@ -98,8 +105,8 @@ async function seedTrackedResources(
 /**
  * managed-agent：help / 缺参不依赖密钥；所有 mutation 命令的 --dry-run
  * 必须在构建 SDK runtime（凭证注入 / 联网 / 写盘）之前短路，因此同样不需要密钥。
- * 鉴权分层：离线命令（init/validate/state list|show|rm）auth: "none"；联网命令
- * 统一 auth: "apiKey" 硬门禁（见 managed-agent-auth-chain e2e）。
+ * 鉴权分层：离线命令（init/project init/validate/state list|show|rm）auth: "none"；
+ * 联网命令统一 auth: "apiKey" 硬门禁（见 managed-agent-auth-chain e2e）。
  * 真实集成（apply/destroy/session 流程）依赖工作区内的 agents.yaml 与远端资源，
  * 属批量场景，暂仅覆盖 dry-run 契约。
  */
@@ -226,6 +233,215 @@ describe("e2e: managed-agent", () => {
     expect(exitCode, stderr).toBe(0);
     expect(stderr).toMatch(/--file|--yes/i);
     expect(stderr).not.toContain("--provider");
+    expect(stderr).not.toContain("--ci");
+    expect(stderr).not.toContain("--refresh-only");
+  });
+
+  test("managed-agent apply 不再接受 --refresh-only", async () => {
+    const { stderr, exitCode } = await runCommandE2e(MANAGED_AGENT_ROUTES, [
+      "managed-agent",
+      "apply",
+      "--dry-run",
+      "--refresh-only",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/Unknown flag.*--refresh-only/i);
+  });
+
+  test("managed-agent init 不再暴露 Git 仓库脚手架", async () => {
+    const { stderr, exitCode } = await runCommandE2e(MANAGED_AGENT_ROUTES, [
+      "managed-agent",
+      "init",
+      "--help",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    expect(stderr).not.toContain("--git");
+  });
+
+  test("managed-agent project 暴露目录项目与共享版本管理子命令", async () => {
+    const { stderr, exitCode } = await runCommandE2e(MANAGED_AGENT_ROUTES, [
+      "managed-agent",
+      "project",
+      "--help",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    expect(stderr).toMatch(/init|validate|build|publish|workbench|version/i);
+  });
+
+  test("managed-agent project 完成 init、validate、build 与 version status 本地闭环", async () => {
+    const projectRoot = await mkdtemp(join(tmpdir(), "bailian-managed-agent-project-"));
+    projectDirectories.push(projectRoot);
+    const initialized = await runCommandE2e(MANAGED_AGENT_ROUTES, [
+      "managed-agent",
+      "project",
+      "init",
+      "--project",
+      projectRoot,
+      "--output",
+      "json",
+    ]);
+    expect(initialized.exitCode, initialized.stderr).toBe(0);
+    expect(
+      parseStdoutJson<{ baseline_version?: string }>(initialized.stdout).baseline_version,
+    ).toMatch(/^[a-f0-9]{64}$/);
+    expect(JSON.parse(await readFile(join(projectRoot, "project.json"), "utf8"))).toEqual({
+      version: "1",
+    });
+    expect(await stat(join(projectRoot, ".env")).catch(() => null)).toBeNull();
+    expect(await stat(join(projectRoot, ".gitignore")).catch(() => null)).toBeNull();
+
+    expect(
+      JSON.parse(await readFile(join(projectRoot, "agents/assistant/agent.json"), "utf8")),
+    ).toEqual({ name: "Assistant", model: "qwen3.7-max" });
+    for (const relativePath of [
+      "skills/_examples/example-skill/skill.json",
+      "skills/_examples/example-skill/SKILL.md",
+      "files/_examples/example-file/file.json",
+      "files/_examples/example-file/example.md",
+      "vaults/_examples/example-vault/vault.json",
+      "environments/_examples/example-env/environment.json",
+    ]) {
+      expect((await stat(join(projectRoot, "agents/assistant", relativePath))).isFile()).toBe(true);
+      expect(
+        await readFile(join(projectRoot, "agents/assistant", relativePath, "../README.md"), "utf8"),
+      ).toContain("_examples/");
+    }
+
+    const skillDirectory = join(projectRoot, "agents/assistant/skills/writer");
+    const filesDirectory = join(projectRoot, "agents/assistant/files");
+    const nestedFileDirectory = join(filesDirectory, "mount");
+    await mkdir(skillDirectory, { recursive: true });
+    await mkdir(nestedFileDirectory, { recursive: true });
+    await writeFile(join(skillDirectory, "SKILL.md"), "# Writer\n", "utf8");
+    await writeFile(join(filesDirectory, "brief.txt"), "Build this brief.\n", "utf8");
+    await writeFile(join(nestedFileDirectory, "mount.md"), "Mount this file.\n", "utf8");
+
+    const validated = await runCommandE2e(MANAGED_AGENT_ROUTES, [
+      "managed-agent",
+      "project",
+      "validate",
+      "--project",
+      projectRoot,
+      "--output",
+      "json",
+    ]);
+    expect(validated.exitCode, validated.stderr).toBe(0);
+    expect(
+      parseStdoutJson<{ diagnostics?: Array<{ severity?: string }> }>(validated.stdout).diagnostics,
+    ).not.toContainEqual(expect.objectContaining({ severity: "error" }));
+
+    const vaultDirectory = join(projectRoot, "agents/assistant/vaults/secrets");
+    await mkdir(vaultDirectory, { recursive: true });
+    await writeFile(
+      join(vaultDirectory, "vault.json"),
+      JSON.stringify({
+        id: "secrets",
+        display_name: "Secrets",
+        credentials: [
+          {
+            name: "service",
+            type: "environment_variable",
+            secret_name: "SERVICE_TOKEN",
+            secret_value: "e2e-vault-private-value",
+          },
+        ],
+      }),
+    );
+    const dryBuild = await runCommandE2e(MANAGED_AGENT_ROUTES, [
+      "managed-agent",
+      "project",
+      "build",
+      "--project",
+      projectRoot,
+      "--dry-run",
+      "--output",
+      "json",
+    ]);
+    expect(dryBuild.exitCode, dryBuild.stderr).toBe(0);
+    expect(dryBuild.stdout + dryBuild.stderr).not.toContain("e2e-vault-private-value");
+    expect(dryBuild.stdout).toContain("AGENTS_VAULT_");
+    expect(await stat(join(projectRoot, ".env")).catch(() => null)).toBeNull();
+
+    const built = await runCommandE2e(MANAGED_AGENT_ROUTES, [
+      "managed-agent",
+      "project",
+      "build",
+      "--project",
+      projectRoot,
+      "--output",
+      "json",
+    ]);
+    expect(built.exitCode, built.stderr).toBe(0);
+    const generatedYaml = await readFile(
+      join(projectRoot, ".openagentpack/build/agents.yaml"),
+      "utf8",
+    );
+    expect(generatedYaml).not.toContain("example-");
+    expect(generatedYaml).not.toContain("_examples");
+    expect(generatedYaml).not.toContain("$" + "{SERVICE_TOKEN}");
+    expect(built.stdout + built.stderr).not.toContain("e2e-vault-private-value");
+    expect(await readFile(join(projectRoot, ".env"), "utf8")).toContain("e2e-vault-private-value");
+    expect(await readFile(join(vaultDirectory, "vault.json"), "utf8")).not.toContain(
+      "e2e-vault-private-value",
+    );
+    expect(
+      await readFile(join(projectRoot, ".openagentpack/build/agents.yaml"), "utf8"),
+    ).not.toContain("e2e-vault-private-value");
+    expect(await readFile(join(projectRoot, ".openagentpack/build/agents.yaml"), "utf8")).toContain(
+      "assistant",
+    );
+    expect(await readFile(join(projectRoot, ".openagentpack/build/agents.yaml"), "utf8")).toContain(
+      "provider: bailian",
+    );
+    expect(JSON.parse(await readFile(join(skillDirectory, "skill.json"), "utf8"))).toEqual({
+      id: "writer",
+    });
+    expect(JSON.parse(await readFile(join(filesDirectory, "brief/file.json"), "utf8"))).toEqual({
+      id: "brief",
+      name: "brief.txt",
+      source: "./brief.txt",
+    });
+    expect(JSON.parse(await readFile(join(nestedFileDirectory, "file.json"), "utf8"))).toEqual({
+      id: "mount",
+      name: "mount.md",
+      source: "./mount.md",
+    });
+    const builtAgent = JSON.parse(
+      await readFile(join(projectRoot, "agents/assistant/agent.json"), "utf8"),
+    );
+    expect(builtAgent.skills).toEqual(["writer"]);
+    expect(builtAgent.files).toEqual(
+      expect.arrayContaining([
+        { file: "brief", mount_path: "/mnt/brief.txt" },
+        { file: "mount", mount_path: "/mnt/mount.md" },
+      ]),
+    );
+
+    const status = await runCommandE2e(MANAGED_AGENT_ROUTES, [
+      "managed-agent",
+      "project",
+      "version",
+      "status",
+      "--project",
+      projectRoot,
+      "--output",
+      "json",
+    ]);
+    expect(status.exitCode, status.stderr).toBe(0);
+    expect(parseStdoutJson<{ enabled?: boolean }>(status.stdout).enabled).toBe(true);
+    expect(await stat(join(projectRoot, ".openagentpack/state.json")).catch(() => null)).toBeNull();
+  });
+
+  test("managed-agent project version preview 缺少 --version-id 时退出为用法错误 (2)", async () => {
+    const { stderr, exitCode } = await runCommandE2e(MANAGED_AGENT_ROUTES, [
+      "managed-agent",
+      "project",
+      "version",
+      "preview",
+      "--quiet",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/--version-id|Missing required/i);
   });
 
   test("managed-agent apply 计划失败时在最终错误中保留首条诊断和剩余数量", async () => {
@@ -1289,6 +1505,29 @@ describe("e2e: managed-agent（--dry-run 短路，不联网不写盘）", () =>
     expect(data.provider).toBe("bailian");
   });
 
+  test("project workbench --dry-run 仅输出目录项目启动计划", async () => {
+    const { stdout, stderr, exitCode } = await runCommandE2e(MANAGED_AGENT_ROUTES, [
+      "managed-agent",
+      "project",
+      "workbench",
+      "--dry-run",
+      "--project",
+      "./agent-project",
+      "--no-open",
+      "--output",
+      "json",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    const data = parseStdoutJson<{
+      would_launch?: string;
+      project_root?: string;
+      port?: number;
+    }>(stdout);
+    expect(data.would_launch).toBe("workbench");
+    expect(data.project_root).toBe("./agent-project");
+    expect(data.port).toBe(4848);
+  });
+
   test("init 不再接受 --provider", async () => {
     const { stderr, exitCode } = await runCommandE2e(MANAGED_AGENT_ROUTES, [
       "managed-agent",
@@ -1301,6 +1540,27 @@ describe("e2e: managed-agent（--dry-run 短路，不联网不写盘）", () =>
     expect(stderr).toMatch(/Unknown flag.*--provider/i);
   });
 
+  test("project init --dry-run 不需要密钥且不创建目录", async () => {
+    const parent = await mkdtemp(join(tmpdir(), "bailian-managed-agent-project-dry-run-"));
+    projectDirectories.push(parent);
+    const projectRoot = join(parent, "project");
+    const { stdout, stderr, exitCode } = await runCommandE2e(MANAGED_AGENT_ROUTES, [
+      "managed-agent",
+      "project",
+      "init",
+      "--dry-run",
+      "--project",
+      projectRoot,
+      "--output",
+      "json",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    expect(
+      parseStdoutJson<{ would_initialize_project?: string }>(stdout).would_initialize_project,
+    ).toBe(projectRoot);
+    expect(await stat(projectRoot).catch(() => null)).toBeNull();
+  });
+
   test("apply --dry-run 仅输出计划", async () => {
     const { stdout, stderr, exitCode } = await runCommandE2e(MANAGED_AGENT_ROUTES, [
       "managed-agent",

packages/commands/tests/e2e/pipeline.e2e.test.ts

@@ -240,4 +240,75 @@ describe("e2e: pipeline", () => {
     expect(stdout).toBe("");
     expect(stderr).toMatch(/--events must be one of: jsonl/i);
   });
+
+  test("pipeline image/generate dry-run 继承 Profile watermark=false", async () => {
+    const configDir = await mkdtemp(join(tmpdir(), "bl-pipeline-wm-"));
+    const workflowPath = join(configDir, "image-generate.json");
+    try {
+      await writeFile(
+        join(configDir, "config.json"),
+        JSON.stringify({ api_key: "sk-test-placeholder", watermark: false }, null, 2) + "\n",
+      );
+      await writeFile(
+        workflowPath,
+        JSON.stringify({
+          version: "workflow/v1",
+          steps: [{ id: "gen", type: "image/generate", input: { prompt: "A cat" } }],
+        }),
+      );
+      const { stdout, stderr, exitCode } = await runCommandE2e(
+        PIPELINE_ROUTES,
+        ["pipeline", "run", "--file", workflowPath, "--dry-run", "--output", "json"],
+        { BAILIAN_CONFIG_DIR: configDir, DASHSCOPE_API_KEY: "", DASHSCOPE_BASE_URL: "" },
+      );
+      expect(exitCode, stderr).toBe(0);
+      const report = parseStdoutJson<{
+        status?: string;
+        steps?: Array<{ type?: string; input?: { watermark?: boolean; prompt?: string } }>;
+      }>(stdout);
+      expect(report.status).toBe("planned");
+      expect(report.steps?.[0]).toMatchObject({
+        type: "image/generate",
+        input: { prompt: "A cat", watermark: false },
+      });
+    } finally {
+      await rm(configDir, { recursive: true, force: true });
+    }
+  });
+
+  test("pipeline 步骤显式 watermark=true 覆盖 Profile false", async () => {
+    const configDir = await mkdtemp(join(tmpdir(), "bl-pipeline-wm-ov-"));
+    const workflowPath = join(configDir, "image-generate.json");
+    try {
+      await writeFile(
+        join(configDir, "config.json"),
+        JSON.stringify({ api_key: "sk-test-placeholder", watermark: false }, null, 2) + "\n",
+      );
+      await writeFile(
+        workflowPath,
+        JSON.stringify({
+          version: "workflow/v1",
+          steps: [
+            {
+              id: "gen",
+              type: "image/generate",
+              input: { prompt: "A cat", watermark: true },
+            },
+          ],
+        }),
+      );
+      const { stdout, stderr, exitCode } = await runCommandE2e(
+        PIPELINE_ROUTES,
+        ["pipeline", "run", "--file", workflowPath, "--dry-run", "--output", "json"],
+        { BAILIAN_CONFIG_DIR: configDir, DASHSCOPE_API_KEY: "", DASHSCOPE_BASE_URL: "" },
+      );
+      expect(exitCode, stderr).toBe(0);
+      const report = parseStdoutJson<{
+        steps?: Array<{ input?: { watermark?: boolean } }>;
+      }>(stdout);
+      expect(report.steps?.[0]?.input?.watermark).toBe(true);
+    } finally {
+      await rm(configDir, { recursive: true, force: true });
+    }
+  });
 });

packages/commands/tests/e2e/speech-recognize.e2e.test.ts

@@ -40,11 +40,18 @@ describe("e2e: speech recognize", () => {
           language_hints?: string[];
           language?: string;
           vocabulary_id?: string;
+          vocabulary?: Record<string, number>;
         };
         input?: {
           file_url?: string;
           file_urls?: string[];
-          messages?: Array<{ content?: Array<{ type?: string }> }>;
+          context?: Array<{
+            role?: string;
+            content?: Array<{ type?: string; text?: string }>;
+          }>;
+          messages?: Array<{
+            content?: Array<{ type?: string; text?: string; input_audio?: { data?: string } }>;
+          }>;
         };
       };
     }>(stdout);
@@ -120,6 +127,116 @@ describe("e2e: speech recognize", () => {
     expect(body.request?.input?.messages?.[0]?.content?.[0]?.type).toBe("input_audio");
   });
 
+  test("speech recognize async dry-run 注入 input.context 与 parameters.vocabulary", async () => {
+    const body = await runRecognizeDryRun([
+      "--model",
+      "qwen-audio-3.0-asr-flash-filetrans",
+      "--url",
+      "https://example.com/audio.mp3",
+      "--vocabulary",
+      '{"奋斗者":4,"鲸落":4}',
+      "--context",
+      "奋斗者号 鲸落 深海勇士",
+    ]);
+    expect(body.mode).toBe("async");
+    expect(body.request?.parameters?.vocabulary).toEqual({ 奋斗者: 4, 鲸落: 4 });
+    expect(body.request?.input?.context).toEqual([
+      {
+        role: "user",
+        content: [{ type: "input_text", text: "奋斗者号 鲸落 深海勇士" }],
+      },
+    ]);
+  });
+
+  test("speech recognize sync input-audio dry-run 将 context 前置且 input_audio 在最后", async () => {
+    const body = await runRecognizeDryRun([
+      "--model",
+      "qwen-audio-3.0-asr-flash",
+      "--url",
+      "https://example.com/audio.wav",
+      "--vocabulary",
+      '{"奋斗者":4}',
+      "--context",
+      "奋斗者号",
+    ]);
+    expect(body.mode).toBe("sync");
+    expect(body.request?.parameters?.vocabulary).toEqual({ 奋斗者: 4 });
+    const messages = body.request?.input?.messages ?? [];
+    expect(messages).toHaveLength(2);
+    expect(messages[0]?.content?.[0]).toMatchObject({ type: "input_text", text: "奋斗者号" });
+    expect(messages[1]?.content?.[0]?.type).toBe("input_audio");
+  });
+
+  test("speech recognize 非法 --vocabulary JSON 返回用法错误", async () => {
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "recognize",
+      "--model",
+      "qwen-audio-3.0-asr-flash-filetrans",
+      "--url",
+      "https://example.com/a.wav",
+      "--vocabulary",
+      "{bad json",
+      "--dry-run",
+      "--quiet",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/not valid JSON|--vocabulary/i);
+  });
+
+  test("speech recognize 空 --vocabulary 返回用法错误", async () => {
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "recognize",
+      "--model",
+      "qwen-audio-3.0-asr-flash-filetrans",
+      "--url",
+      "https://example.com/a.wav",
+      "--vocabulary",
+      "",
+      "--dry-run",
+      "--quiet",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/not valid JSON|--vocabulary/i);
+  });
+
+  test("speech recognize qwen3 sync 拒绝 --vocabulary", async () => {
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "recognize",
+      "--model",
+      "qwen3-asr-flash",
+      "--url",
+      "https://example.com/a.wav",
+      "--vocabulary",
+      '{"奋斗者":4}',
+      "--dry-run",
+      "--quiet",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/--vocabulary|does not support/i);
+    expect(stderr).toMatch(/qwen-audio-3\.0-asr-flash|vocabulary\/context/i);
+  });
+
+  test("speech recognize sync Flash 拒绝 --diarization 时提示 async filetrans", async () => {
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "recognize",
+      "--model",
+      "fun-asr-flash",
+      "--url",
+      "https://example.com/a.wav",
+      "--diarization",
+      "--dry-run",
+      "--quiet",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/--diarization|does not support/i);
+    expect(stderr).toMatch(/async filetrans|fun-asr/i);
+    expect(stderr).not.toMatch(/vocabulary\/context/i);
+  });
+
   test("speech recognize qwen3 filetrans dry-run 使用 file_url 与 language", async () => {
     const body = await runRecognizeDryRun([
       "--model",
@@ -366,5 +483,88 @@ describe.skipIf(!isBailianE2EMediaEnabled() || !isDashScopeE2EReady())(
       const raw = readFileSync(asrJson, "utf8");
       expect(raw.length).toBeGreaterThan(2);
     }, 300_000);
+
+    test("【qwen-audio】synthesize → recognize 即时热词/上下文", async () => {
+      // 生造专名：无热词时常被听错；带 --vocabulary/--context 后应能正确召回。
+      const hotwordScript =
+        "请把录音同步到听悟匣，并启动澜舟芯做摘要。听悟匣负责转写，澜舟芯负责归档。最后确认玄甲协议是否已开启。";
+      const hotwords = ["听悟匣", "澜舟芯", "玄甲协议"] as const;
+      const vocabularyJson = '{"听悟匣":4,"澜舟芯":4,"玄甲协议":4}';
+      const contextText = "听悟匣 澜舟芯 玄甲协议";
+
+      const missingHotwords = (text: string): string[] => {
+        const normalized = text.replace(/\s+/g, "");
+        return hotwords.filter((word) => !normalized.includes(word));
+      };
+
+      const outDir = makeE2eOutputDir(e2eLabelFromMetaUrl(import.meta.url));
+      const outMp3 = join(outDir, "hotword-tts.mp3");
+      const syn = await runCommandE2e(SPEECH_ROUTES, [
+        "speech",
+        "synthesize",
+        "--model",
+        "cosyvoice-v3-flash",
+        "--voice",
+        "longxiaochun_v3",
+        "--text",
+        hotwordScript,
+        "--out",
+        outMp3,
+        "--output",
+        "json",
+      ]);
+      expect(syn.exitCode, syn.stderr).toBe(0);
+      const synBody = parseStdoutJson<{ audio_url?: string }>(syn.stdout);
+      const audioUrl = synBody.audio_url;
+      expect(audioUrl?.startsWith("http")).toBe(true);
+
+      const baselineOut = join(outDir, "asr-baseline.json");
+      const baseline = await runCommandE2e(SPEECH_ROUTES, [
+        "speech",
+        "recognize",
+        "--model",
+        "qwen-audio-3.0-asr-flash",
+        "--url",
+        audioUrl!,
+        "--language",
+        "zh",
+        "--out",
+        baselineOut,
+        "--quiet",
+      ]);
+      expect(baseline.exitCode, baseline.stderr).toBe(0);
+      writeFileSync(join(outDir, "asr-baseline.txt"), baseline.stdout);
+      const baselineMissing = missingHotwords(baseline.stdout);
+      // soft：仅落盘对照，不 fail（无热词偶发也能认出专名）
+      writeFileSync(
+        join(outDir, "asr-baseline-missing.txt"),
+        baselineMissing.length > 0 ? baselineMissing.join("\n") + "\n" : "(none)\n",
+      );
+
+      const hotOut = join(outDir, "asr-hot.json");
+      const hot = await runCommandE2e(SPEECH_ROUTES, [
+        "speech",
+        "recognize",
+        "--model",
+        "qwen-audio-3.0-asr-flash",
+        "--url",
+        audioUrl!,
+        "--language",
+        "zh",
+        "--vocabulary",
+        vocabularyJson,
+        "--context",
+        contextText,
+        "--out",
+        hotOut,
+        "--quiet",
+      ]);
+      expect(hot.exitCode, hot.stderr).toBe(0);
+      writeFileSync(join(outDir, "asr-hot.txt"), hot.stdout);
+      expect(
+        missingHotwords(hot.stdout),
+        `expected hotwords in ASR text, got: ${hot.stdout.trim()}`,
+      ).toEqual([]);
+    }, 420_000);
   },
 );

packages/commands/tests/e2e/speech-vocabulary.e2e.test.ts

@@ -0,0 +1,551 @@
+import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, describe, expect, test } from "vite-plus/test";
+import {
+  e2eLabelFromMetaUrl,
+  isBailianE2EMediaEnabled,
+  isDashScopeE2EReady,
+  makeE2eOutputDir,
+  parseStdoutJson,
+  runCommandHelp,
+  runCommandE2e,
+} from "./helpers.ts";
+import { SPEECH_ROUTES } from "./topic-routes.ts";
+
+/**
+ * Speech vocabulary：help / dry-run / 确认闸门无密钥；真实 CRUD 需媒体 E2E + DashScope。
+ */
+
+const tempDirs: string[] = [];
+
+afterEach(() => {
+  for (const tempDir of tempDirs.splice(0)) {
+    rmSync(tempDir, { recursive: true, force: true });
+  }
+});
+
+function makeTempJson(content: string): string {
+  const tempDir = mkdtempSync(join(tmpdir(), "bl-vocab-e2e-"));
+  tempDirs.push(tempDir);
+  const filePath = join(tempDir, "hotwords.json");
+  writeFileSync(filePath, content);
+  return filePath;
+}
+
+describe("e2e: speech vocabulary", () => {
+  test("speech vocabulary --help 列出 5 个子命令", async () => {
+    const { stderr, exitCode } = await runCommandHelp(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "--help",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    expect(stderr).toMatch(/create/i);
+    expect(stderr).toMatch(/list/i);
+    expect(stderr).toMatch(/get/i);
+    expect(stderr).toMatch(/update/i);
+    expect(stderr).toMatch(/delete/i);
+  });
+
+  test("create --help 展示关键 flags 与静默失效 / weight 50 文案", async () => {
+    const { stderr, exitCode } = await runCommandHelp(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "create",
+      "--help",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    expect(stderr).toMatch(/--model/i);
+    expect(stderr).toMatch(/--prefix/i);
+    expect(stderr).toMatch(/--words/i);
+    expect(stderr).toMatch(/--words-file/i);
+    expect(stderr).toMatch(/silently ignored|静默失效/i);
+    expect(stderr).toMatch(/50/);
+  });
+
+  test("delete --help 展示 Risk 与 --yes", async () => {
+    const { stderr, exitCode } = await runCommandHelp(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "delete",
+      "--help",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    expect(stderr).toMatch(/--yes/i);
+    expect(stderr).toMatch(/Risk|风险/i);
+  });
+
+  test("update --help 展示 Risk、replace 语义与 --yes", async () => {
+    const { stderr, exitCode } = await runCommandHelp(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "update",
+      "--help",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    expect(stderr).toMatch(/--yes/i);
+    expect(stderr).toMatch(/Risk|风险/i);
+    expect(stderr).toMatch(/replaces|替换/i);
+  });
+
+  test("create 缺少 --model 时退出为用法错误 (2)", async () => {
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "create",
+      "--prefix",
+      "demo",
+      "--words",
+      '{"x":4}',
+      "--quiet",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/--model|Missing required/i);
+  });
+
+  test("create 缺少 --prefix 时退出为用法错误 (2)", async () => {
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "create",
+      "--model",
+      "fun-asr",
+      "--words",
+      '{"x":4}',
+      "--quiet",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/--prefix|Missing required/i);
+  });
+
+  test("create 两个词表 flag 都不传时退出为用法错误 (2)", async () => {
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "create",
+      "--model",
+      "fun-asr",
+      "--prefix",
+      "demo",
+      "--quiet",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/--words|--words-file/i);
+  });
+
+  test("create 同时传 --words 与 --words-file 时退出为用法错误 (2)", async () => {
+    const filePath = makeTempJson('{"x":4}');
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "create",
+      "--model",
+      "fun-asr",
+      "--prefix",
+      "demo",
+      "--words",
+      '{"x":4}',
+      "--words-file",
+      filePath,
+      "--quiet",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/either|--words|--words-file/i);
+  });
+
+  test("create 非法 JSON 时退出为用法错误 (2)", async () => {
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "create",
+      "--model",
+      "fun-asr",
+      "--prefix",
+      "demo",
+      "--words",
+      "{bad json",
+      "--quiet",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/not valid JSON|JSON/i);
+  });
+
+  test("create 字符串权重时退出为用法错误 (2)", async () => {
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "create",
+      "--model",
+      "fun-asr",
+      "--prefix",
+      "demo",
+      "--words",
+      '{"x":"4"}',
+      "--quiet",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/must be a number|number/i);
+  });
+
+  test("create 空词表 {} 时退出为用法错误 (2)", async () => {
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "create",
+      "--model",
+      "fun-asr",
+      "--prefix",
+      "demo",
+      "--words",
+      "{}",
+      "--quiet",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/at least one|hot word/i);
+  });
+
+  test("get 缺少 --id 时退出为用法错误 (2)", async () => {
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "get",
+      "--quiet",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/--id|Missing required/i);
+  });
+
+  test("delete 缺少 --id 时退出为用法错误 (2)", async () => {
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "delete",
+      "--quiet",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toMatch(/--id|Missing required/i);
+  });
+
+  test("create object 形态 --dry-run 输出 speech-biasing 信封", async () => {
+    const { stdout, stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "create",
+      "--model",
+      "fun-asr",
+      "--prefix",
+      "demo",
+      "--words",
+      '{"奋斗者":4,"鲸落":4}',
+      "--dry-run",
+      "--output",
+      "json",
+      "--quiet",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    const body = parseStdoutJson<{
+      request?: {
+        model?: string;
+        input?: {
+          action?: string;
+          target_model?: string;
+          prefix?: string;
+          vocabulary?: Array<{ text?: string; weight?: number; lang?: string }>;
+        };
+      };
+    }>(stdout);
+    expect(body.request?.model).toBe("speech-biasing");
+    expect(body.request?.input?.action).toBe("create_vocabulary");
+    expect(body.request?.input?.target_model).toBe("fun-asr");
+    expect(body.request?.input?.prefix).toBe("demo");
+    expect(body.request?.input?.vocabulary).toEqual([
+      { text: "奋斗者", weight: 4 },
+      { text: "鲸落", weight: 4 },
+    ]);
+  });
+
+  test("create array 形态 --dry-run 原样透传 lang", async () => {
+    const { stdout, stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "create",
+      "--model",
+      "fun-asr",
+      "--prefix",
+      "demo",
+      "--words",
+      '[{"text":"x","weight":4,"lang":"zh"}]',
+      "--dry-run",
+      "--output",
+      "json",
+      "--quiet",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    const body = parseStdoutJson<{
+      request?: {
+        input?: { vocabulary?: Array<{ text?: string; weight?: number; lang?: string }> };
+      };
+    }>(stdout);
+    expect(body.request?.input?.vocabulary).toEqual([{ text: "x", weight: 4, lang: "zh" }]);
+  });
+
+  test("create object + --lang --dry-run 下发到每一条", async () => {
+    const { stdout, stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "create",
+      "--model",
+      "fun-asr",
+      "--prefix",
+      "demo",
+      "--words",
+      '{"x":4}',
+      "--lang",
+      "zh",
+      "--dry-run",
+      "--output",
+      "json",
+      "--quiet",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    const body = parseStdoutJson<{
+      request?: {
+        input?: { vocabulary?: Array<{ text?: string; weight?: number; lang?: string }> };
+      };
+    }>(stdout);
+    expect(body.request?.input?.vocabulary).toEqual([{ text: "x", weight: 4, lang: "zh" }]);
+  });
+
+  test("create --words-file --dry-run 读取文件", async () => {
+    const filePath = makeTempJson('{"from-file":4}');
+    const { stdout, stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "create",
+      "--model",
+      "fun-asr",
+      "--prefix",
+      "demo",
+      "--words-file",
+      filePath,
+      "--dry-run",
+      "--output",
+      "json",
+      "--quiet",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    const body = parseStdoutJson<{
+      request?: {
+        input?: { vocabulary?: Array<{ text?: string; weight?: number }> };
+      };
+    }>(stdout);
+    expect(body.request?.input?.vocabulary).toEqual([{ text: "from-file", weight: 4 }]);
+  });
+
+  test("list --page 2 --dry-run 将 page_index 转为 1", async () => {
+    const { stdout, stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "list",
+      "--page",
+      "2",
+      "--dry-run",
+      "--output",
+      "json",
+      "--quiet",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    const body = parseStdoutJson<{
+      request?: { input?: { action?: string; page_index?: number } };
+    }>(stdout);
+    expect(body.request?.input?.action).toBe("list_vocabulary");
+    expect(body.request?.input?.page_index).toBe(1);
+  });
+
+  test("get --dry-run 使用 query_vocabulary", async () => {
+    const { stdout, stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "get",
+      "--id",
+      "vocab-x",
+      "--dry-run",
+      "--output",
+      "json",
+      "--quiet",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    const body = parseStdoutJson<{
+      request?: { input?: { action?: string; vocabulary_id?: string } };
+    }>(stdout);
+    expect(body.request?.input?.action).toBe("query_vocabulary");
+    expect(body.request?.input?.vocabulary_id).toBe("vocab-x");
+  });
+
+  test("update --dry-run 无 --yes 也能预览", async () => {
+    const { stdout, stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "update",
+      "--id",
+      "vocab-x",
+      "--words",
+      '{"x":4}',
+      "--dry-run",
+      "--output",
+      "json",
+      "--quiet",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    const body = parseStdoutJson<{
+      request?: {
+        input?: {
+          action?: string;
+          vocabulary_id?: string;
+          vocabulary?: Array<{ text?: string; weight?: number }>;
+        };
+      };
+    }>(stdout);
+    expect(body.request?.input?.action).toBe("update_vocabulary");
+    expect(body.request?.input?.vocabulary_id).toBe("vocab-x");
+    expect(body.request?.input?.vocabulary).toEqual([{ text: "x", weight: 4 }]);
+  });
+
+  test("delete --dry-run 无 --yes 也能预览", async () => {
+    const { stdout, stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "delete",
+      "--id",
+      "vocab-x",
+      "--dry-run",
+      "--output",
+      "json",
+      "--quiet",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    const body = parseStdoutJson<{
+      request?: { input?: { action?: string; vocabulary_id?: string } };
+    }>(stdout);
+    expect(body.request?.input?.action).toBe("delete_vocabulary");
+    expect(body.request?.input?.vocabulary_id).toBe("vocab-x");
+  });
+});
+
+describe("e2e: speech vocabulary high-risk confirmation", () => {
+  test("delete 无 --yes 返回确认请求 (7)", async () => {
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "delete",
+      "--id",
+      "vocab-x",
+      "--api-key",
+      "e2e-dummy-key",
+      "--output",
+      "json",
+    ]);
+    expect(exitCode).toBe(7);
+    expect(JSON.parse(stderr)).toMatchObject({
+      error: { code: 7, type: "requires_confirmation" },
+    });
+  });
+
+  test("update 无 --yes 返回确认请求 (7)", async () => {
+    const { stderr, exitCode } = await runCommandE2e(SPEECH_ROUTES, [
+      "speech",
+      "vocabulary",
+      "update",
+      "--id",
+      "vocab-x",
+      "--words",
+      '{"x":4}',
+      "--api-key",
+      "e2e-dummy-key",
+      "--output",
+      "json",
+    ]);
+    expect(exitCode).toBe(7);
+    expect(JSON.parse(stderr)).toMatchObject({
+      error: { code: 7, type: "requires_confirmation" },
+    });
+  });
+});
+
+describe.skipIf(!isBailianE2EMediaEnabled() || !isDashScopeE2EReady())(
+  "e2e: speech vocabulary（DashScope 媒体）",
+  () => {
+    test("create → list/get → delete 完整链路", async () => {
+      const outDir = makeE2eOutputDir(e2eLabelFromMetaUrl(import.meta.url));
+      let vocabularyId = "";
+
+      try {
+        const created = await runCommandE2e(SPEECH_ROUTES, [
+          "speech",
+          "vocabulary",
+          "create",
+          "--model",
+          "fun-asr",
+          "--prefix",
+          "blcli",
+          "--words",
+          '{"奋斗者":4}',
+          "--quiet",
+        ]);
+        expect(created.exitCode, created.stderr).toBe(0);
+        vocabularyId = created.stdout.trim();
+        expect(vocabularyId.length).toBeGreaterThan(0);
+        writeFileSync(join(outDir, "vocabulary-id.txt"), vocabularyId + "\n");
+
+        const listed = await runCommandE2e(SPEECH_ROUTES, [
+          "speech",
+          "vocabulary",
+          "list",
+          "--prefix",
+          "blcli",
+          "--output",
+          "json",
+        ]);
+        expect(listed.exitCode, listed.stderr).toBe(0);
+        const listBody = parseStdoutJson<{
+          output?: { vocabulary_list?: Array<{ vocabulary_id?: string; status?: string }> };
+        }>(listed.stdout);
+        const listedItem = listBody.output?.vocabulary_list?.find(
+          (item) => item.vocabulary_id === vocabularyId,
+        );
+        expect(listedItem).toBeTruthy();
+        expect(listedItem?.status).toBe("OK");
+
+        const got = await runCommandE2e(SPEECH_ROUTES, [
+          "speech",
+          "vocabulary",
+          "get",
+          "--id",
+          vocabularyId,
+          "--output",
+          "json",
+        ]);
+        expect(got.exitCode, got.stderr).toBe(0);
+        const getBody = parseStdoutJson<{
+          output?: { status?: string; target_model?: string };
+        }>(got.stdout);
+        expect(getBody.output?.status).toBe("OK");
+        expect(getBody.output?.target_model).toBe("fun-asr");
+      } finally {
+        if (vocabularyId) {
+          const deleted = await runCommandE2e(SPEECH_ROUTES, [
+            "speech",
+            "vocabulary",
+            "delete",
+            "--id",
+            vocabularyId,
+            "--yes",
+            "--quiet",
+          ]);
+          expect(deleted.exitCode, deleted.stderr).toBe(0);
+        }
+      }
+    }, 120_000);
+  },
+);

packages/commands/tests/e2e/topic-routes.ts

@@ -72,6 +72,11 @@ export const VISION_ROUTES: E2eRouteExports = {
 export const SPEECH_ROUTES: E2eRouteExports = {
   "speech synthesize": "speechSynthesize",
   "speech recognize": "speechRecognize",
+  "speech vocabulary create": "speechVocabularyCreate",
+  "speech vocabulary list": "speechVocabularyList",
+  "speech vocabulary get": "speechVocabularyGet",
+  "speech vocabulary update": "speechVocabularyUpdate",
+  "speech vocabulary delete": "speechVocabularyDelete",
 };
 
 export const MCP_ROUTES: E2eRouteExports = {
@@ -188,6 +193,18 @@ export const MANAGED_AGENT_ROUTES: E2eRouteExports = {
   "managed-agent plan": "managedAgentPlan",
   "managed-agent apply": "managedAgentApply",
   "managed-agent destroy": "managedAgentDestroy",
+  "managed-agent playground": "managedAgentPlayground",
+  "managed-agent project init": "managedAgentProjectInit",
+  "managed-agent project validate": "managedAgentProjectValidate",
+  "managed-agent project build": "managedAgentProjectBuild",
+  "managed-agent project publish": "managedAgentProjectPublish",
+  "managed-agent project workbench": "managedAgentProjectWorkbench",
+  "managed-agent project version enable": "managedAgentProjectVersionEnable",
+  "managed-agent project version disable": "managedAgentProjectVersionDisable",
+  "managed-agent project version status": "managedAgentProjectVersionStatus",
+  "managed-agent project version list": "managedAgentProjectVersionList",
+  "managed-agent project version preview": "managedAgentProjectVersionPreview",
+  "managed-agent project version restore": "managedAgentProjectVersionRestore",
   "managed-agent state list": "managedAgentStateList",
   "managed-agent state rm": "managedAgentStateRm",
   "managed-agent state import": "managedAgentStateImport",

packages/commands/tests/e2e/watermark-config.e2e.test.ts

@@ -0,0 +1,80 @@
+import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { describe, expect, test } from "vite-plus/test";
+import { parseStdoutJson, runCommandE2e } from "./helpers.ts";
+import { IMAGE_ROUTES, VIDEO_ROUTES, type E2eRouteExports } from "./topic-routes.ts";
+
+interface WatermarkScenario {
+  name: string;
+  routes: E2eRouteExports;
+  args: string[];
+}
+
+const scenarios: WatermarkScenario[] = [
+  {
+    name: "image generate",
+    routes: IMAGE_ROUTES,
+    args: ["image", "generate", "--prompt", "A cat"],
+  },
+  {
+    name: "image edit",
+    routes: IMAGE_ROUTES,
+    args: [
+      "image",
+      "edit",
+      "--image",
+      "https://example.com/input.png",
+      "--prompt",
+      "Blue background",
+    ],
+  },
+  {
+    name: "video generate",
+    routes: VIDEO_ROUTES,
+    args: ["video", "generate", "--prompt", "A cat waves"],
+  },
+  {
+    name: "video edit",
+    routes: VIDEO_ROUTES,
+    args: ["video", "edit", "--video", "https://example.com/input.mp4", "--prompt", "Warm colors"],
+  },
+  {
+    name: "video ref",
+    routes: VIDEO_ROUTES,
+    args: [
+      "video",
+      "ref",
+      "--image",
+      "https://example.com/person.png",
+      "--prompt",
+      "Image 1 waves",
+    ],
+  },
+];
+
+describe("e2e: global watermark config", () => {
+  for (const scenario of scenarios) {
+    test(`${scenario.name} uses watermark=false from the selected Profile`, async () => {
+      const configDir = mkdtempSync(join(tmpdir(), "bl-watermark-profile-"));
+      try {
+        writeFileSync(
+          join(configDir, "config.json"),
+          JSON.stringify({ media: { watermark: false } }, null, 2) + "\n",
+        );
+        const { stdout, stderr, exitCode } = await runCommandE2e(
+          scenario.routes,
+          [...scenario.args, "--config", "media", "--dry-run", "--output", "json"],
+          { BAILIAN_CONFIG_DIR: configDir },
+        );
+        expect(exitCode, stderr).toBe(0);
+        const data = parseStdoutJson<{
+          request?: { parameters?: { watermark?: boolean } };
+        }>(stdout);
+        expect(data.request?.parameters?.watermark).toBe(false);
+      } finally {
+        rmSync(configDir, { recursive: true, force: true });
+      }
+    });
+  }
+});

packages/commands/tests/managed-agent-local-versions.test.ts

@@ -0,0 +1,139 @@
+import { chmod, mkdir, mkdtemp, readFile, rm, stat, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import {
+  commitPreparedProjectVersion,
+  disableProjectVersioning,
+  enableProjectVersioning,
+  getProjectVersionStatus,
+  listProjectVersions,
+  prepareProjectVersion,
+  previewProjectVersion,
+  restoreProjectVersion,
+} from "@openagentpack/sdk/project-versions";
+import { afterEach, describe, expect, test } from "vite-plus/test";
+import { playgroundBrowserTargetFromSummary } from "../src/commands/managed-agent/_engine/playground-launcher.ts";
+
+const temporaryDirectories: string[] = [];
+
+afterEach(async () => {
+  for (const directory of temporaryDirectories.splice(0)) {
+    await rm(directory, { recursive: true, force: true });
+  }
+});
+
+describe("managed-agent local snapshot versions", () => {
+  test("uses the shared path-scoped switch and stores full YAML outside store.json", async () => {
+    const root = await temporaryDirectory();
+    const configPath = join(root, "agents.yaml");
+    const siblingDirectory = join(root, "nested");
+    const siblingConfigPath = join(siblingDirectory, "agents.yaml");
+    await mkdir(siblingDirectory);
+    await writeFile(configPath, projectYaml("First"));
+    await writeFile(siblingConfigPath, projectYaml("Second"));
+
+    const enabled = await enableProjectVersioning(configPath, "Enable Bailian CLI versioning");
+
+    expect(enabled.versioning.enabled).toBe(true);
+    expect(enabled.version?.message).toBe("Enable Bailian CLI versioning");
+    expect((await getProjectVersionStatus(siblingConfigPath)).enabled).toBe(false);
+    const storeSource = await readFile(join(root, ".openagentpack/versions/store.json"), "utf8");
+    expect(storeSource).not.toContain("instructions: First");
+    const snapshotSource = await readFile(
+      join(root, ".openagentpack/versions/blobs", `${enabled.version!.source_hash}.yaml`),
+      "utf8",
+    );
+    expect(snapshotSource).toContain("instructions: First");
+
+    await writeFile(configPath, projectYaml("First updated"));
+    const repeated = await enableProjectVersioning(configPath, "Enable Bailian CLI versioning");
+    expect(repeated.version?.message).toBe("Enable Bailian CLI versioning");
+    expect((await listProjectVersions(configPath)).versions).toHaveLength(2);
+
+    const disabled = await disableProjectVersioning(configPath);
+    expect(disabled.enabled).toBe(false);
+  });
+
+  test("auto-snapshots after success and restores without changing history or permissions", async () => {
+    const root = await temporaryDirectory();
+    const configPath = join(root, "agents.yaml");
+    await writeFile(configPath, projectYaml("Version one"));
+    await chmod(configPath, 0o640);
+    const enabled = await enableProjectVersioning(configPath, "Enable Bailian CLI versioning");
+    const firstVersion = enabled.version!.version_id;
+    const secondSource = projectYaml("Version two");
+    await writeFile(configPath, secondSource);
+
+    const prepared = await prepareProjectVersion(configPath, secondSource);
+    const version = await commitPreparedProjectVersion(prepared!);
+    const currentVersionBeforeRestore = (await getProjectVersionStatus(configPath)).head_version;
+    expect(version?.message).toBe("Apply agents.yaml");
+
+    const preview = await previewProjectVersion(configPath, firstVersion);
+    expect(preview.can_restore).toBe(true);
+    expect(preview.after_yaml).toContain("Version one");
+    await restoreProjectVersion(configPath, firstVersion, {
+      headVersion: preview.base_head_version,
+      sourceRevision: preview.base_source_revision,
+    });
+
+    expect(await readFile(configPath, "utf8")).toContain("Version one");
+    expect((await getProjectVersionStatus(configPath)).head_version).toBe(
+      currentVersionBeforeRestore,
+    );
+    expect((await stat(configPath)).mode & 0o777).toBe(0o640);
+  });
+
+  test("rejects abbreviated version IDs and plaintext credentials", async () => {
+    const root = await temporaryDirectory();
+    const configPath = join(root, "agents.yaml");
+    await writeFile(configPath, projectYaml("Safe"));
+    const enabled = await enableProjectVersioning(configPath, "Enable Bailian CLI versioning");
+    await expect(previewProjectVersion(configPath, enabled.version!.short_version)).rejects.toThrow(
+      /full 64-character hexadecimal/i,
+    );
+
+    await disableProjectVersioning(configPath);
+    await writeFile(
+      configPath,
+      projectYaml("Unsafe").replace("qoder: {}", "qoder:\n    api_key: plaintext-secret"),
+    );
+    await expect(
+      enableProjectVersioning(configPath, "Enable Bailian CLI versioning"),
+    ).rejects.toThrow(/environment variable reference/i);
+  });
+});
+
+describe("managed-agent Workbench policy", () => {
+  test("Session Preview opens the requested Agent or falls back to Workbench", () => {
+    const summary = {
+      status: "valid",
+      agents: [{ agent: { id: "assistant" } }, { agent: { id: "reviewer" } }],
+    };
+    expect(
+      playgroundBrowserTargetFromSummary("http://localhost:4848", summary, "reviewer"),
+    ).toEqual({ url: "http://localhost:4848/agents/reviewer/preview" });
+    expect(playgroundBrowserTargetFromSummary("http://localhost:4848", summary)).toEqual(
+      expect.objectContaining({ url: "http://localhost:4848", warning: expect.any(String) }),
+    );
+  });
+});
+
+async function temporaryDirectory(): Promise<string> {
+  const directory = await mkdtemp(join(tmpdir(), "bailian-cli-local-versions-"));
+  temporaryDirectories.push(directory);
+  return directory;
+}
+
+function projectYaml(instructions: string): string {
+  return `version: "1"
+providers:
+  qoder: {}
+defaults:
+  provider: qoder
+agents:
+  assistant:
+    model: ultimate
+    instructions: ${instructions}
+`;
+}

packages/commands/tests/managed-agent-playground-launcher.test.ts

@@ -0,0 +1,137 @@
+import { mkdir, mkdtemp, rm, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, beforeEach, expect, test, vi } from "vite-plus/test";
+
+const registry = vi.hoisted(() => ({
+  version: '"0.7.0"',
+  error: null as Error | null,
+  manifest: "",
+  query: vi.fn(),
+}));
+
+vi.mock("node:child_process", async (importOriginal) => ({
+  ...(await importOriginal<typeof import("node:child_process")>()),
+  execFile: (
+    command: string,
+    args: string[],
+    options: unknown,
+    callback: (error: Error | null, output: { stdout: string }) => void,
+  ) => {
+    registry.query(command, args, options);
+    callback(registry.error, { stdout: registry.version });
+  },
+}));
+
+vi.mock("node:module", () => ({
+  createRequire: () => ({ resolve: () => registry.manifest }),
+}));
+
+import { resolveLauncher } from "../src/commands/managed-agent/_engine/playground-launcher.ts";
+
+let directory: string;
+let binary: string;
+
+beforeEach(async () => {
+  directory = await mkdtemp(join(tmpdir(), "playground-launcher-"));
+  binary = join(directory, "playground.js");
+  registry.manifest = join(directory, "package.json");
+  registry.version = '"0.7.0"';
+  registry.error = null;
+  registry.query.mockClear();
+  await writeFile(binary, "");
+  vi.stubEnv("BAILIAN_MANAGED_AGENT_PLAYGROUND_BIN", "");
+  vi.stubEnv("AGENTS_PLAYGROUND_BIN", "");
+  vi.stubEnv("BAILIAN_MANAGED_AGENT_PLAYGROUND_VERSION", "");
+  vi.spyOn(process, "cwd").mockReturnValue(directory);
+});
+
+afterEach(async () => {
+  vi.unstubAllEnvs();
+  vi.restoreAllMocks();
+  await rm(directory, { recursive: true, force: true });
+});
+
+async function install(version: string) {
+  await writeFile(registry.manifest, JSON.stringify({ version, bin: "playground.js" }));
+}
+
+test("checks latest on every resolution and reuses only the matching installed version", async () => {
+  await install("0.7.0");
+  expect(await resolveLauncher()).toMatchObject({
+    args: [binary],
+    version: "0.7.0",
+    fetched: false,
+  });
+  registry.version = '"0.8.0"';
+  expect(await resolveLauncher()).toMatchObject({
+    command: "npx",
+    args: ["-y", "@openagentpack/playground@0.8.0"],
+    version: "0.8.0",
+    fetched: true,
+  });
+  expect(registry.query).toHaveBeenCalledTimes(2);
+  expect(registry.query).toHaveBeenCalledWith(
+    "npm",
+    expect.arrayContaining([
+      "view",
+      "@openagentpack/playground@latest",
+      "--prefer-online",
+      "--fetch-timeout=10000",
+    ]),
+    expect.objectContaining({ timeout: 15_000 }),
+  );
+});
+
+test("downloads the resolved exact version when local installation is old or missing", async () => {
+  for (const version of ["0.6.0", undefined]) {
+    if (version) await install(version);
+    else await rm(registry.manifest);
+    expect(await resolveLauncher()).toMatchObject({
+      command: "npx",
+      args: ["-y", "@openagentpack/playground@0.7.0"],
+    });
+  }
+});
+
+test("explicit version overrides an incompatible installed version", async () => {
+  await install("0.6.0");
+  vi.stubEnv("BAILIAN_MANAGED_AGENT_PLAYGROUND_VERSION", "0.7.0");
+  expect(await resolveLauncher()).toMatchObject({ fetched: true, version: "0.7.0" });
+  expect(registry.query.mock.calls[0]?.[1]).toContain("@openagentpack/playground@0.7.0");
+});
+
+test("explicit binary remains an offline development override", async () => {
+  vi.stubEnv("BAILIAN_MANAGED_AGENT_PLAYGROUND_BIN", binary);
+  expect(await resolveLauncher()).toMatchObject({ args: [binary], fetched: false });
+  expect(registry.query).not.toHaveBeenCalled();
+});
+
+test("registry failure does not silently reuse an old installation", async () => {
+  await install("0.6.0");
+  registry.error = new Error("registry unavailable");
+  await expect(resolveLauncher()).rejects.toThrow("registry unavailable");
+});
+
+test("rejects malformed or ambiguous registry metadata with localized diagnostics", async () => {
+  for (const output of ["invalid", '["0.6.0","0.7.0"]', '"--unsafe"']) {
+    registry.version = output;
+    await expect(resolveLauncher()).rejects.toThrow("single valid Playground version");
+    await expect(resolveLauncher()).rejects.toThrow("唯一有效的 Playground 版本号");
+  }
+});
+
+test("local source builds must also match the resolved version", async () => {
+  const packageRoot = join(directory, "packages/playground");
+  const sourceBinary = join(packageRoot, "dist/bin/playground.js");
+  await mkdir(join(packageRoot, "dist/bin"), { recursive: true });
+  await writeFile(sourceBinary, "");
+  await writeFile(join(packageRoot, "package.json"), JSON.stringify({ version: "0.6.0" }));
+  expect(await resolveLauncher()).toMatchObject({ fetched: true });
+  await writeFile(join(packageRoot, "package.json"), JSON.stringify({ version: "0.7.0" }));
+  expect(await resolveLauncher()).toMatchObject({
+    args: [sourceBinary],
+    version: "0.7.0",
+    fetched: false,
+  });
+});

packages/core/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bailian-cli-core",
-  "version": "1.20.0",
+  "version": "1.23.0",
   "description": "Core SDK for bailian-cli. See https://www.npmjs.com/package/bailian-cli for usage.",
   "homepage": "https://bailian.console.aliyun.com/cli",
   "bugs": {

packages/core/src/client/asr-routes.ts

@@ -1,16 +1,20 @@
 import { imageSyncPath, speechRecognizePath } from "./endpoints.ts";
 
+import type { AsrContextMessage } from "../types/api.ts";
+
 /**
  * DashScope ASR APIs differ by model family:
  *
  * - async file transcription (`.../audio/asr/transcription`):
  *   fun-asr*, paraformer* (non-realtime), *-filetrans, sensevoice*
- *   language via `parameters.language_hints`
+ *   language via `parameters.language_hints`; optional `input.context` /
+ *   `parameters.vocabulary` (model-dependent effective range)
  * - sync multimodal (`.../aigc/multimodal-generation/generation`):
  *   - qwen3: `{ content: [{ audio }] }` + optional `asr_options.language`
- *     (qwen3-asr-flash*)
+ *     (qwen3-asr-flash*) — no vocabulary / context fields in this body shape
  *   - input-audio: `{ type: input_audio, input_audio.data }` +
- *     `format`/`sample_rate` + optional `language_hints`
+ *     `format`/`sample_rate` + optional `language_hints` /
+ *     `vocabulary_id` / `vocabulary`; optional leading `input_text` for context
  *     (fun-asr-flash*, qwen-audio-*-asr-flash*)
  * - realtime / streaming: WebSocket — not supported by `speech recognize`
  */
@@ -158,9 +162,18 @@ export interface BuildAsrFlashRequestOpts {
   language?: string;
   /** Precompiled hotword vocabulary ID; supported for input-audio Flash (fun-asr-flash* / qwen-audio-*-asr-flash). */
   vocabularyId?: string;
+  /** Instant hot words (word → weight); input-audio Flash only (command layer rejects qwen3). */
+  vocabulary?: Record<string, number>;
+  /** Context enhancement text; prepended as input_text before input_audio. */
+  context?: string;
   flashFamily: AsrFlashFamily;
 }
 
+/** Wrap plain text as a single user context message for ASR. */
+export function buildAsrContextMessages(text: string): AsrContextMessage[] {
+  return [{ role: "user", content: [{ type: "input_text", text }] }];
+}
+
 /**
  * Build language fields for async ASR routes.
  * qwen3-asr-flash-filetrans* → `language`; other async models → `language_hints`.
@@ -178,10 +191,10 @@ export function buildAsyncAsrLanguageFields(
 
 /** Build a sync multimodal ASR request body for Flash models. */
 export function buildAsrFlashRequest(opts: BuildAsrFlashRequestOpts): Record<string, unknown> {
-  const { model, audioUrl, language, vocabularyId, flashFamily } = opts;
+  const { model, audioUrl, language, vocabularyId, vocabulary, context, flashFamily } = opts;
 
   if (flashFamily === "input-audio") {
-    // Match official Qwen-Audio / Fun-ASR-Flash docs: language_hints + vocabulary_id
+    // Match official Qwen-Audio / Fun-ASR-Flash docs: language_hints + vocabulary(_id)
     const parameters: Record<string, unknown> = {
       format: inferAudioFormatHint(audioUrl),
       sample_rate: "16000",
@@ -192,21 +205,28 @@ export function buildAsrFlashRequest(opts: BuildAsrFlashRequestOpts): Record<str
     if (vocabularyId) {
       parameters.vocabulary_id = vocabularyId;
     }
+    if (vocabulary) {
+      parameters.vocabulary = vocabulary;
+    }
+    // input_audio must be the last message; prepend context as input_text when present
+    const messages: Array<Record<string, unknown>> = [];
+    if (context) {
+      for (const message of buildAsrContextMessages(context)) {
+        messages.push(message as unknown as Record<string, unknown>);
+      }
+    }
+    messages.push({
+      role: "user",
+      content: [
+        {
+          type: "input_audio",
+          input_audio: { data: audioUrl },
+        },
+      ],
+    });
     return {
       model,
-      input: {
-        messages: [
-          {
-            role: "user",
-            content: [
-              {
-                type: "input_audio",
-                input_audio: { data: audioUrl },
-              },
-            ],
-          },
-        ],
-      },
+      input: { messages },
       parameters,
     };
   }

packages/core/src/client/endpoints.ts

@@ -89,6 +89,11 @@ export function speechRecognizePath(): string {
   return "/api/v1/services/audio/asr/transcription";
 }
 
+// ---- Hot-word Vocabulary (ASR customization) ----
+export function speechVocabularyPath(): string {
+  return "/api/v1/services/audio/asr/customization";
+}
+
 // ---- Memory Profile (DashScope v2) ----
 export function profileSchemaPath(): string {
   return "/api/v2/apps/memory/profile_schemas";

packages/core/src/client/index.ts

@@ -21,6 +21,7 @@ export {
   responsesPath,
   speechRecognizePath,
   speechSynthesizePath,
+  speechVocabularyPath,
   taskPath,
   userProfilePath,
   videoGeneratePath,
@@ -41,6 +42,7 @@ export {
   type ImageSizeProfile,
 } from "./image-routes.ts";
 export {
+  buildAsrContextMessages,
   buildAsrFlashRequest,
   buildAsyncAsrLanguageFields,
   collectAsrTranscriptionItems,

packages/core/src/config/loader.ts

@@ -272,6 +272,7 @@ export function buildSettings(s: ResolutionSources): Settings {
     outputExplicit: Boolean(flags.output || env.DASHSCOPE_OUTPUT || file.output),
     outputDir: file.output_dir || undefined,
     timeout,
+    watermark: file.watermark ?? true,
     defaultTextModel: file.default_text_model,
     defaultVideoModel: file.default_video_model,
     defaultImageToVideoModel: file.default_image_to_video_model,

packages/core/src/config/schema.ts

@@ -35,6 +35,7 @@ export interface ConfigFile {
   output?: "text" | "json";
   output_dir?: string;
   timeout?: number;
+  watermark?: boolean;
   default_text_model?: string;
   default_video_model?: string;
   default_image_to_video_model?: string;
@@ -63,6 +64,7 @@ export const CONFIG_FILE_KEYS = [
   "output",
   "output_dir",
   "timeout",
+  "watermark",
   "default_text_model",
   "default_video_model",
   "default_image_to_video_model",
@@ -156,6 +158,7 @@ export function parseConfigFile(raw: unknown): ConfigFile {
   if (typeof obj.output_dir === "string" && obj.output_dir.length > 0)
     out.output_dir = obj.output_dir;
   if (typeof obj.timeout === "number" && obj.timeout > 0) out.timeout = obj.timeout;
+  if (typeof obj.watermark === "boolean") out.watermark = obj.watermark;
   if (typeof obj.default_text_model === "string" && obj.default_text_model.length > 0)
     out.default_text_model = obj.default_text_model;
   if (typeof obj.default_video_model === "string" && obj.default_video_model.length > 0)
@@ -224,6 +227,7 @@ export interface Settings {
   outputExplicit: boolean;
   outputDir?: string;
   timeout: number;
+  watermark: boolean;
   defaultTextModel?: string;
   defaultVideoModel?: string;
   defaultImageToVideoModel?: string;

packages/core/src/index.ts

@@ -13,6 +13,7 @@ export * from "./files/index.ts";
 export * from "./dataset/index.ts";
 export * from "./finetune/index.ts";
 export * from "./deploy/index.ts";
+export * from "./speech/index.ts";
 export * from "./types/index.ts";
 export * from "./utils/index.ts";
 export * from "./telemetry/index.ts";

packages/core/src/speech/index.ts

@@ -0,0 +1,14 @@
+export {
+  SPEECH_BIASING_MODEL,
+  buildVocabularyRequest,
+  createVocabulary,
+  listVocabularies,
+  queryVocabulary,
+  updateVocabulary,
+  deleteVocabulary,
+  type VocabularyEntry,
+  type VocabularyListItem,
+  type VocabularyEnvelope,
+  type VocabularyRequest,
+} from "./vocabulary.ts";
+export { parseInstantVocabulary, parseVocabularyEntries } from "./vocabulary-input.ts";

packages/core/src/speech/vocabulary-input.ts

@@ -0,0 +1,76 @@
+import { UsageError } from "../errors/base.ts";
+import type { VocabularyEntry } from "./vocabulary.ts";
+
+/** Shared JSON decode + top-level shape guard for both hot-word flags. */
+function decodeVocabularyJson(raw: string, flagName: string): unknown {
+  try {
+    return JSON.parse(raw);
+  } catch (error) {
+    throw new UsageError(`${flagName} is not valid JSON — ${(error as Error).message}`);
+  }
+}
+
+/**
+ * Instant hot words for `recognize --vocabulary`.
+ * The API field is a flat word→weight object, so an array is a usage error here.
+ */
+export function parseInstantVocabulary(raw: string): Record<string, number> {
+  const parsed = decodeVocabularyJson(raw, "--vocabulary");
+  if (!parsed || typeof parsed !== "object" || Array.isArray(parsed)) {
+    throw new UsageError("--vocabulary must decode to a JSON object of word→weight.");
+  }
+  for (const [word, weight] of Object.entries(parsed)) {
+    if (typeof weight !== "number" || !Number.isFinite(weight)) {
+      throw new UsageError(`--vocabulary weight for "${word}" must be a number.`);
+    }
+  }
+  return parsed as Record<string, number>;
+}
+
+/**
+ * Vocabulary create/update --words.
+ * Accepts the same word→weight object as recognize, or the API entry array
+ * when per-entry lang is needed. Array form ignores the optional lang param.
+ */
+export function parseVocabularyEntries(raw: string, lang?: string): VocabularyEntry[] {
+  const flagName = "--words";
+  const parsed = decodeVocabularyJson(raw, flagName);
+  let entries: VocabularyEntry[];
+  if (Array.isArray(parsed)) {
+    entries = parsed.map((item, index) => {
+      if (!item || typeof item !== "object" || Array.isArray(item)) {
+        throw new UsageError(
+          `${flagName} entry #${index} must be an object with string "text" and number "weight".`,
+        );
+      }
+      const entry = item as Partial<VocabularyEntry>;
+      if (typeof entry.text !== "string" || typeof entry.weight !== "number") {
+        throw new UsageError(
+          `${flagName} entry #${index} must have a string "text" and a number "weight".`,
+        );
+      }
+      if (!Number.isFinite(entry.weight)) {
+        throw new UsageError(`${flagName} entry #${index} weight must be a finite number.`);
+      }
+      return {
+        text: entry.text,
+        weight: entry.weight,
+        ...(typeof entry.lang === "string" ? { lang: entry.lang } : {}),
+      };
+    });
+  } else if (!parsed || typeof parsed !== "object") {
+    throw new UsageError(`${flagName} must decode to a JSON object or array.`);
+  } else {
+    entries = [];
+    for (const [text, weight] of Object.entries(parsed)) {
+      if (typeof weight !== "number" || !Number.isFinite(weight)) {
+        throw new UsageError(`${flagName} weight for "${text}" must be a number.`);
+      }
+      entries.push({ text, weight, ...(lang ? { lang } : {}) });
+    }
+  }
+  if (entries.length === 0) {
+    throw new UsageError(`${flagName} must contain at least one hot word.`);
+  }
+  return entries;
+}

packages/core/src/speech/vocabulary.ts

@@ -0,0 +1,128 @@
+/**
+ * Hot-word vocabulary HTTP API wrappers.
+ *
+ * Thin functions over `requestJson`. They return the parsed body verbatim
+ * (snake_case) so callers can decide how to surface fields.
+ */
+import { speechVocabularyPath } from "../client/endpoints.ts";
+import type { Client } from "../client/client.ts";
+
+/** Fixed model id for the hot-word customization endpoint. */
+export const SPEECH_BIASING_MODEL = "speech-biasing";
+
+export interface VocabularyEntry {
+  text: string;
+  weight: number;
+  lang?: string;
+}
+
+export interface VocabularyListItem {
+  vocabulary_id?: string;
+  gmt_create?: string;
+  gmt_modified?: string;
+  /** OK | UNDEPLOYED — UNDEPLOYED vocabularies are silently ignored by ASR. */
+  status?: string;
+}
+
+export interface VocabularyEnvelope<T> {
+  request_id?: string;
+  output?: T;
+  usage?: { count?: number };
+}
+
+export interface VocabularyRequest {
+  model: string;
+  input: Record<string, unknown>;
+}
+
+/** Shared request body builder for dry-run and live calls. */
+export function buildVocabularyRequest(
+  action: string,
+  input: Record<string, unknown>,
+): VocabularyRequest {
+  return {
+    model: SPEECH_BIASING_MODEL,
+    input: { action, ...input },
+  };
+}
+
+async function callVocabularyApi<T>(
+  client: Client,
+  action: string,
+  input: Record<string, unknown>,
+  signal?: AbortSignal,
+): Promise<VocabularyEnvelope<T>> {
+  return client.requestJson<VocabularyEnvelope<T>>({
+    path: speechVocabularyPath(),
+    method: "POST",
+    body: buildVocabularyRequest(action, input),
+    signal,
+  });
+}
+
+export function createVocabulary(
+  client: Client,
+  params: { targetModel: string; prefix: string; vocabulary: VocabularyEntry[] },
+  signal?: AbortSignal,
+): Promise<VocabularyEnvelope<{ vocabulary_id?: string }>> {
+  return callVocabularyApi(
+    client,
+    "create_vocabulary",
+    {
+      target_model: params.targetModel,
+      prefix: params.prefix,
+      vocabulary: params.vocabulary,
+    },
+    signal,
+  );
+}
+
+export function listVocabularies(
+  client: Client,
+  params: { prefix?: string; pageIndex?: number; pageSize?: number } = {},
+  signal?: AbortSignal,
+): Promise<VocabularyEnvelope<{ vocabulary_list?: VocabularyListItem[] }>> {
+  const input: Record<string, unknown> = {};
+  if (params.prefix !== undefined) input.prefix = params.prefix;
+  if (params.pageIndex !== undefined) input.page_index = params.pageIndex;
+  if (params.pageSize !== undefined) input.page_size = params.pageSize;
+  return callVocabularyApi(client, "list_vocabulary", input, signal);
+}
+
+export function queryVocabulary(
+  client: Client,
+  vocabularyId: string,
+  signal?: AbortSignal,
+): Promise<
+  VocabularyEnvelope<{
+    gmt_create?: string;
+    gmt_modified?: string;
+    status?: string;
+    target_model?: string;
+    vocabulary?: VocabularyEntry[];
+  }>
+> {
+  return callVocabularyApi(client, "query_vocabulary", { vocabulary_id: vocabularyId }, signal);
+}
+
+export function updateVocabulary(
+  client: Client,
+  vocabularyId: string,
+  vocabulary: VocabularyEntry[],
+  signal?: AbortSignal,
+): Promise<VocabularyEnvelope<Record<string, never>>> {
+  return callVocabularyApi(
+    client,
+    "update_vocabulary",
+    { vocabulary_id: vocabularyId, vocabulary },
+    signal,
+  );
+}
+
+export function deleteVocabulary(
+  client: Client,
+  vocabularyId: string,
+  signal?: AbortSignal,
+): Promise<VocabularyEnvelope<Record<string, never>>> {
+  return callVocabularyApi(client, "delete_vocabulary", { vocabulary_id: vocabularyId }, signal);
+}

packages/core/src/types/api.ts

@@ -580,11 +580,19 @@ export interface DashScopeTTSStreamChunk {
 
 // ---- Speech Recognition / ASR (DashScope) ----
 
+/** Context-enhancement message for async ASR `input.context` / sync Flash `input.messages`. */
+export interface AsrContextMessage {
+  role: "user" | "assistant";
+  content: Array<{ type: "input_text" | "text"; text: string }>;
+}
+
 export interface DashScopeASRRequest {
   model: string;
   input: {
     file_urls?: string[];
     file_url?: string;
+    /** Context enhancement for async filetrans (array of chat-style messages). */
+    context?: AsrContextMessage[];
   };
   parameters?: {
     channel_id?: number[];
@@ -595,6 +603,8 @@ export interface DashScopeASRRequest {
     diarization_enabled?: boolean;
     speaker_count?: number;
     vocabulary_id?: string;
+    /** Instant hot words (word → weight); takes effect on Qwen-Audio-3.0-ASR-Flash series. */
+    vocabulary?: Record<string, number>;
   };
 }
 

packages/core/src/types/command.ts

@@ -287,8 +287,10 @@ export interface Command<F extends FlagsDef = FlagsDef> {
    * Cross-flag validation, after parsing and before run. Return an error message
    * → UsageError; undefined to pass. Single-flag `required` is enforced by the
    * parser — use this for rules spanning flags or depending on a flag's *value*.
+   * May be async for read-only local preflight. Runtime awaits it before auth
+   * and confirmation. Do not perform remote requests or local writes here.
    */
-  validate?: (flags: ParsedFlags<F>) => string | undefined;
+  validate?: (flags: ParsedFlags<F>) => string | undefined | Promise<string | undefined>;
   run: (ctx: CommandContext<F>) => Promise<void>;
 }
 

packages/core/src/types/index.ts

@@ -49,6 +49,7 @@ export type {
   ChatRequest,
   ChatResponse,
   ChatTool,
+  AsrContextMessage,
   DashScopeASRRequest,
   DashScopeASRTaskResult,
   DashScopeASRTranscriptionItem,

packages/core/src/utils/boolean-flag.ts

@@ -34,7 +34,7 @@ export function resolveBooleanFlag(
   return defaultWhenUnset;
 }
 
-/** Resolve `--watermark` flag; default true when unset. */
-export function resolveWatermark(flagValue: unknown): boolean {
-  return parseOptionalBooleanValue(flagValue, "watermark") ?? true;
+/** Resolve `--watermark`; command flag overrides config, then defaults to true. */
+export function resolveWatermark(flagValue: unknown, configuredValue?: boolean): boolean {
+  return parseOptionalBooleanValue(flagValue, "watermark") ?? configuredValue ?? true;
 }

packages/core/tests/asr-routes.test.ts

@@ -1,5 +1,6 @@
 import { expect, test } from "vite-plus/test";
 import {
+  buildAsrContextMessages,
   buildAsrFlashRequest,
   buildAsyncAsrLanguageFields,
   collectAsrTranscriptionItems,
@@ -141,6 +142,47 @@ test("buildAsrFlashRequest shapes qwen3 and input-audio bodies", () => {
   });
 });
 
+test("buildAsrContextMessages wraps plain text as a single user input_text message", () => {
+  expect(buildAsrContextMessages("奋斗者号 鲸落")).toEqual([
+    { role: "user", content: [{ type: "input_text", text: "奋斗者号 鲸落" }] },
+  ]);
+});
+
+test("buildAsrFlashRequest injects instant vocabulary and prepends context before input_audio", () => {
+  const body = buildAsrFlashRequest({
+    model: "qwen-audio-3.0-asr-flash",
+    audioUrl: "https://example.com/a.wav",
+    vocabulary: { 奋斗者: 4, 鲸落: 4 },
+    context: "奋斗者号 鲸落",
+    flashFamily: "input-audio",
+  });
+
+  expect(body).toEqual({
+    model: "qwen-audio-3.0-asr-flash",
+    input: {
+      messages: [
+        {
+          role: "user",
+          content: [{ type: "input_text", text: "奋斗者号 鲸落" }],
+        },
+        {
+          role: "user",
+          content: [{ type: "input_audio", input_audio: { data: "https://example.com/a.wav" } }],
+        },
+      ],
+    },
+    parameters: {
+      format: "wav",
+      sample_rate: "16000",
+      vocabulary: { 奋斗者: 4, 鲸落: 4 },
+    },
+  });
+
+  const messages = (body.input as { messages: Array<{ content: Array<{ type?: string }> }> })
+    .messages;
+  expect(messages[messages.length - 1]?.content?.[0]?.type).toBe("input_audio");
+});
+
 test("buildAsyncAsrLanguageFields maps language by async style", () => {
   expect(buildAsyncAsrLanguageFields("language_hints", "zh")).toEqual({
     language_hints: ["zh"],

packages/core/tests/config-priority.test.ts

@@ -79,6 +79,14 @@ test("default_speech_recognition_model 从配置文件进入运行时 Settings",
   expect(resolve({ file }).defaultSpeechRecognitionModel).toBe("qwen-audio-3.0-asr-flash");
 });
 
+test("watermark 从配置文件进入 Settings，缺省时保持合规默认值 true", () => {
+  expect(parseConfigFile({ watermark: false }).watermark).toBe(false);
+  expect(parseConfigFile({ watermark: true }).watermark).toBe(true);
+  expect(parseConfigFile({ watermark: "false" }).watermark).toBeUndefined();
+  expect(resolve({ file: { watermark: false } }).watermark).toBe(false);
+  expect(resolve({}).watermark).toBe(true);
+});
+
 test("baseUrl:flag > env > file > 默认，所有来源统一归一化", () => {
   const flags = { baseUrl: "https://flag.example.com/compatible-mode/v1?source=flag" };
   const env = { DASHSCOPE_BASE_URL: "https://env.example.com/apps/anthropic#env" };

packages/core/tests/image-input.test.ts

@@ -28,6 +28,7 @@ function makeSettings(configName?: string): Settings {
     output: "json",
     outputExplicit: false,
     timeout: 30,
+    watermark: true,
     verbose: false,
     quiet: true,
     dryRun: false,

packages/core/tests/index.test.ts

@@ -31,6 +31,7 @@ function testDeps(identity: Partial<Identity> = {}): {
       output: "json",
       outputExplicit: true,
       timeout: 30,
+      watermark: true,
       verbose: false,
       quiet: true,
       dryRun: false,
@@ -290,6 +291,9 @@ test("resolveWatermark uses flag or defaults to true", () => {
   expect(resolveWatermark("false")).toBe(false);
   expect(resolveWatermark("true")).toBe(true);
   expect(resolveWatermark(undefined)).toBe(true);
+  expect(resolveWatermark(undefined, false)).toBe(false);
+  expect(resolveWatermark("true", false)).toBe(true);
+  expect(resolveWatermark("false", true)).toBe(false);
 });
 
 test("resolveBooleanFlag uses flag or defaultWhenUnset", () => {

packages/core/tests/mcp.test.ts

@@ -23,6 +23,7 @@ function testDeps(overrides?: Partial<Settings>): { identity: Identity; settings
       output: "json",
       outputExplicit: true,
       timeout: 5,
+      watermark: true,
       verbose: false,
       quiet: true,
       dryRun: false,

packages/core/tests/speech-vocabulary-input.test.ts

@@ -0,0 +1,106 @@
+import { describe, expect, test } from "vite-plus/test";
+import {
+  parseInstantVocabulary,
+  parseVocabularyEntries,
+  buildVocabularyRequest,
+  SPEECH_BIASING_MODEL,
+  UsageError,
+} from "../src/index.ts";
+
+describe("parseInstantVocabulary", () => {
+  test("接受 word→weight 对象", () => {
+    expect(parseInstantVocabulary('{"奋斗者":4,"鲸落":5}')).toEqual({
+      奋斗者: 4,
+      鲸落: 5,
+    });
+  });
+
+  test("拒绝数组", () => {
+    expect(() => parseInstantVocabulary('[{"text":"x","weight":4}]')).toThrow(UsageError);
+    expect(() => parseInstantVocabulary('[{"text":"x","weight":4}]')).toThrow(/object/i);
+  });
+
+  test("拒绝字符串权重", () => {
+    expect(() => parseInstantVocabulary('{"x":"4"}')).toThrow(UsageError);
+    expect(() => parseInstantVocabulary('{"x":"4"}')).toThrow(/must be a number/);
+  });
+
+  test("拒绝非有限数字权重（null）", () => {
+    expect(() => parseInstantVocabulary('{"x":null}')).toThrow(UsageError);
+    expect(() => parseInstantVocabulary('{"x":null}')).toThrow(/must be a number/);
+  });
+
+  test("拒绝 NaN / Infinity 字面量（非法 JSON）", () => {
+    expect(() => parseInstantVocabulary('{"x":NaN}')).toThrow(UsageError);
+    expect(() => parseInstantVocabulary('{"x":Infinity}')).toThrow(UsageError);
+  });
+
+  test("拒绝顶层 null", () => {
+    expect(() => parseInstantVocabulary("null")).toThrow(UsageError);
+  });
+
+  test("非法 JSON 抛 UsageError", () => {
+    expect(() => parseInstantVocabulary("{bad json")).toThrow(UsageError);
+    expect(() => parseInstantVocabulary("{bad json")).toThrow(/not valid JSON/);
+  });
+});
+
+describe("parseVocabularyEntries", () => {
+  test("对象形态转为条目数组", () => {
+    expect(parseVocabularyEntries('{"奋斗者":4,"鲸落":5}')).toEqual([
+      { text: "奋斗者", weight: 4 },
+      { text: "鲸落", weight: 5 },
+    ]);
+  });
+
+  test("对象形态叠加 --lang", () => {
+    expect(parseVocabularyEntries('{"奋斗者":4}', "zh")).toEqual([
+      { text: "奋斗者", weight: 4, lang: "zh" },
+    ]);
+  });
+
+  test("数组形态透传 lang", () => {
+    expect(parseVocabularyEntries('[{"text":"奋斗者","weight":4,"lang":"zh"}]')).toEqual([
+      { text: "奋斗者", weight: 4, lang: "zh" },
+    ]);
+  });
+
+  test("数组形态忽略第二参 lang", () => {
+    expect(parseVocabularyEntries('[{"text":"奋斗者","weight":4}]', "zh")).toEqual([
+      { text: "奋斗者", weight: 4 },
+    ]);
+    expect(parseVocabularyEntries('[{"text":"奋斗者","weight":4,"lang":"en"}]', "zh")).toEqual([
+      { text: "奋斗者", weight: 4, lang: "en" },
+    ]);
+  });
+
+  test("数组缺 text 或 weight 拒绝", () => {
+    expect(() => parseVocabularyEntries('[{"weight":4}]')).toThrow(UsageError);
+    expect(() => parseVocabularyEntries('[{"text":"x"}]')).toThrow(UsageError);
+  });
+
+  test("非 object 数组元素拒绝", () => {
+    expect(() => parseVocabularyEntries('["x"]')).toThrow(UsageError);
+    expect(() => parseVocabularyEntries("[null]")).toThrow(UsageError);
+  });
+
+  test("空对象 / 空数组拒绝", () => {
+    expect(() => parseVocabularyEntries("{}")).toThrow(UsageError);
+    expect(() => parseVocabularyEntries("{}")).toThrow(/at least one/);
+    expect(() => parseVocabularyEntries("[]")).toThrow(UsageError);
+    expect(() => parseVocabularyEntries("[]")).toThrow(/at least one/);
+  });
+});
+
+describe("buildVocabularyRequest", () => {
+  test("固定 model 为 speech-biasing", () => {
+    const body = buildVocabularyRequest("create_vocabulary", {
+      target_model: "fun-asr",
+      prefix: "demo",
+      vocabulary: [{ text: "奋斗者", weight: 4 }],
+    });
+    expect(body.model).toBe(SPEECH_BIASING_MODEL);
+    expect(body.model).toBe("speech-biasing");
+    expect(body.input.action).toBe("create_vocabulary");
+  });
+});

packages/kscli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "knowledge-studio-cli",
-  "version": "1.20.0",
+  "version": "1.23.0",
   "description": "Lightweight RAG CLI for Aliyun Model Studio — focused on knowledge-base retrieval.",
   "keywords": [
     "alibaba-cloud",

packages/runtime/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bailian-cli-runtime",
-  "version": "1.20.0",
+  "version": "1.23.0",
   "description": "Runtime framework for bailian-cli (createCli, registry, args, output, pipeline). See https://www.npmjs.com/package/bailian-cli for usage.",
   "homepage": "https://bailian.console.aliyun.com/cli",
   "bugs": {

packages/runtime/src/create-cli.ts

@@ -217,7 +217,7 @@ export function createCli(commands: Record<string, AnyCommand>, opts: CliOptions
             parsedFlags,
             Object.keys(res.command.flags ?? {}),
           ) as ParsedFlags<FlagsDef>;
-          const invalid = res.command.validate?.(ownFlags);
+          const invalid = await res.command.validate?.(ownFlags);
           if (invalid) throw new UsageError(invalid);
 
           // 校验通过 → 建源、解析 settings、组 ctx,进中间件执行命令。

packages/runtime/src/pipeline/bl-config.ts

@@ -16,6 +16,22 @@ export interface PipelineEnv {
   settings: Settings;
 }
 
+/** Media steps that inherit Profile `watermark` when the YAML omits the field. */
+export const PIPELINE_WATERMARK_STEPS = new Set(["image/generate", "image/edit", "video/generate"]);
+
+/**
+ * Fill Profile watermark into a planned/executed step input.
+ * Explicit step `watermark` wins; otherwise use Settings (file → default true).
+ */
+export function applyProfileWatermarkToStepInput(
+  stepType: string,
+  input: Record<string, unknown>,
+  settings: Settings,
+): Record<string, unknown> {
+  if (!PIPELINE_WATERMARK_STEPS.has(stepType) || input.watermark !== undefined) return input;
+  return { ...input, watermark: settings.watermark };
+}
+
 /**
  * Build the in-process env for pipeline steps. Uses the same source resolution
  * as the CLI itself (env vars, config file; no CLI flags), but forces JSON

packages/runtime/src/pipeline/executor.ts

@@ -1,5 +1,5 @@
 import { PipelineError, toPipelineError } from "./errors.ts";
-import { buildPipelineEnv } from "./bl-config.ts";
+import { applyProfileWatermarkToStepInput, buildPipelineEnv } from "./bl-config.ts";
 import { getDefaultStepDispatcher, type StepDispatcher } from "./dispatcher.ts";
 import {
   evaluateCondition,
@@ -156,12 +156,17 @@ async function executePipelineInternal(
   if (options.dryRun) {
     for (const planStep of topologicalOrder(plan)) {
       const resolved = resolvePlannedStepInput(planStep.step, pipeline, normalizedRuntimeInput);
+      const plannedInput = applyProfileWatermarkToStepInput(
+        planStep.step.type,
+        resolved.redacted,
+        blEnv.settings,
+      );
       const report: PipelineStepReport = {
         id: planStep.step.id,
         type: planStep.step.type,
         status: "planned",
         dependencies: planStep.dependencies,
-        input: resolved.redacted,
+        input: plannedInput,
         ...(planStep.step.when !== undefined ? { condition: "pending" } : {}),
       };
       reports.push(report);
@@ -170,14 +175,14 @@ async function executePipelineInternal(
         timestamp: now(),
         status: "planned",
         step: stepEvent(planStep),
-        input: inputSummary(resolved.redacted, resolved.sensitiveKeys),
+        input: inputSummary(plannedInput, resolved.sensitiveKeys),
       });
       await emit({
         type: "step.planned",
         timestamp: now(),
         status: "planned",
         step: stepEvent(planStep),
-        input: inputSummary(resolved.redacted, resolved.sensitiveKeys),
+        input: inputSummary(plannedInput, resolved.sensitiveKeys),
         ...(planStep.step.when !== undefined ? { condition: "pending" as const } : {}),
       });
     }

packages/runtime/src/pipeline/steps/bl-api.ts

@@ -12,6 +12,7 @@ import {
   speechRecognizePath,
   resolveAsrApi,
   buildAsrFlashRequest,
+  buildAsrContextMessages,
   buildAsyncAsrLanguageFields,
   collectAsrTranscriptionItems,
   extractAsrFlashText,
@@ -192,7 +193,8 @@ export async function imageGenerate(
     n,
     seed: input.seed,
     prompt_extend: promptExtend,
-    watermark: resolveWatermark(input.watermark),
+    // Step input overrides Profile; omit → use Profile / CLI default (true).
+    watermark: resolveWatermark(input.watermark, env.settings.watermark),
   };
 
   const body: DashScopeImageRequest =
@@ -299,7 +301,8 @@ export async function imageEdit(
     n,
     seed: input.seed,
     prompt_extend: promptExtend,
-    watermark: resolveWatermark(input.watermark),
+    // Step input overrides Profile; omit → use Profile / CLI default (true).
+    watermark: resolveWatermark(input.watermark, env.settings.watermark),
   };
 
   let body: DashScopeImageRequest;
@@ -460,7 +463,8 @@ export async function videoGenerate(
       ratio: input.ratio || undefined,
       duration: input.duration,
       prompt_extend: resolveBooleanFlag(input["prompt-extend"], undefined, "prompt-extend"),
-      watermark: resolveWatermark(input.watermark),
+      // Step input overrides Profile; omit → use Profile / CLI default (true).
+      watermark: resolveWatermark(input.watermark, env.settings.watermark),
       seed: input.seed,
     },
   };
@@ -563,6 +567,10 @@ export interface SpeechRecognizeInput {
   diarization?: boolean;
   "speaker-count"?: number;
   "vocabulary-id"?: string;
+  /** Instant hot words (already structured; no JSON string parse needed). */
+  vocabulary?: Record<string, number>;
+  /** Context enhancement plain text. */
+  context?: string;
   "channel-id"?: number;
   "poll-interval"?: number;
 }
@@ -602,9 +610,11 @@ export async function speechRecognize(
     const unsupportedFlags: string[] = [];
     if (input.diarization) unsupportedFlags.push("diarization");
     if (input["speaker-count"] !== undefined) unsupportedFlags.push("speaker-count");
-    // input-audio Flash supports vocabulary_id; qwen3 sync Flash does not
-    if (route.flashFamily === "qwen3" && input["vocabulary-id"] !== undefined) {
-      unsupportedFlags.push("vocabulary-id");
+    // qwen3 sync Flash has no place for vocabulary_id / vocabulary / context in its body shape
+    if (route.flashFamily === "qwen3") {
+      if (input["vocabulary-id"] !== undefined) unsupportedFlags.push("vocabulary-id");
+      if (input.vocabulary !== undefined) unsupportedFlags.push("vocabulary");
+      if (input.context !== undefined) unsupportedFlags.push("context");
     }
     if (input["channel-id"] !== undefined) unsupportedFlags.push("channel-id");
     if (unsupportedFlags.length > 0) {
@@ -648,6 +658,8 @@ export async function speechRecognize(
       audioUrl: fileUrls[0]!,
       language: input.language,
       vocabularyId: input["vocabulary-id"],
+      vocabulary: input.vocabulary,
+      context: input.context,
       flashFamily,
     });
     const response = await env.client.requestJson<Record<string, unknown>>({
@@ -671,14 +683,19 @@ export async function speechRecognize(
   );
   const body: DashScopeASRRequest = {
     model,
-    input:
-      route.asyncInputStyle === "file_url" ? { file_url: fileUrls[0]! } : { file_urls: fileUrls },
+    input: {
+      ...(route.asyncInputStyle === "file_url"
+        ? { file_url: fileUrls[0]! }
+        : { file_urls: fileUrls }),
+      ...(input.context ? { context: buildAsrContextMessages(input.context) } : {}),
+    },
     parameters: {
       channel_id: input["channel-id"] !== undefined ? [input["channel-id"]] : undefined,
       ...languageFields,
       diarization_enabled: input.diarization,
       speaker_count: input["speaker-count"],
       vocabulary_id: input["vocabulary-id"],
+      vocabulary: input.vocabulary,
     },
   };
   stripUndefined(body.parameters as Record<string, unknown>);

packages/runtime/src/pipeline/steps/bl-steps.ts

@@ -1,5 +1,9 @@
 import { registerStep } from "../dispatcher.ts";
-import { buildPipelineEnv, type PipelineEnv } from "../bl-config.ts";
+import {
+  applyProfileWatermarkToStepInput,
+  buildPipelineEnv,
+  type PipelineEnv,
+} from "../bl-config.ts";
 import { isRecord } from "../utils.ts";
 import {
   textChat,
@@ -153,9 +157,13 @@ async function executeDirectBlStep(
   input: Record<string, unknown>,
   ctx: StepContext,
 ): Promise<StepResult> {
+  const env = (ctx.blEnv as PipelineEnv | undefined) ?? buildPipelineEnv();
+
   if (ctx.dryRun) {
+    // Surface the Profile-effective watermark in dry-run so plans match runtime.
+    const plannedInput = applyProfileWatermarkToStepInput(id, input, env.settings);
     return {
-      metadata: { dryRun: true, step: id, plannedInput: input },
+      metadata: { dryRun: true, step: id, plannedInput },
       warnings: [
         { code: "dry_run_skipped", message: `Step ${id} was not executed in dry-run mode` },
       ],
@@ -167,7 +175,6 @@ async function executeDirectBlStep(
     throw new Error(`No direct API handler registered for step: ${id}`);
   }
 
-  const env = (ctx.blEnv as PipelineEnv | undefined) ?? buildPipelineEnv();
   const data = await handler(env, input, ctx);
   const builder = RESULT_BUILDERS[id];
   if (builder) {

packages/runtime/src/utils/flag-descriptions.ts

@@ -1,8 +1,9 @@
 /** Shared --foo <bool> help text; keep wording consistent with actual CLI/request behavior. */
 
 export const BOOL_FLAG_WATERMARK = {
-  "en-US": "Enable watermark (true/false). Omit flag to use CLI default (true).",
-  "zh-CN": "是否启用水印（true/false）。不传时使用 CLI 默认值 true。",
+  "en-US":
+    "Enable watermark (true/false). Omit flag to use the Profile setting or CLI default (true).",
+  "zh-CN": "是否启用水印（true/false）。不传时使用 Profile 配置或 CLI 默认值 true。",
 };
 
 /** CLI sends prompt_extend=true when flag omitted (qwen-image edit, etc.). */

packages/runtime/tests/speech-recognize-pipeline.test.ts

@@ -68,6 +68,81 @@ test("pipeline speechRecognize routes input-audio flash to sync multimodal endpo
   });
 });
 
+test("pipeline speechRecognize injects vocabulary and context on sync input-audio", async () => {
+  const { env, captured } = makeEnv();
+  await speechRecognize(
+    env,
+    {
+      url: "https://example.com/a.wav",
+      model: "qwen-audio-3.0-asr-flash",
+      vocabulary: { 奋斗者: 4 },
+      context: "奋斗者号",
+    },
+    makeCtx(),
+  );
+
+  expect(captured[0]?.body).toMatchObject({
+    parameters: { vocabulary: { 奋斗者: 4 } },
+    input: {
+      messages: [
+        { role: "user", content: [{ type: "input_text", text: "奋斗者号" }] },
+        {
+          role: "user",
+          content: [{ type: "input_audio", input_audio: { data: "https://example.com/a.wav" } }],
+        },
+      ],
+    },
+  });
+});
+
+test("pipeline speechRecognize injects vocabulary and context on async filetrans", async () => {
+  const { env, captured } = makeEnv(async (opts) => {
+    if (opts.async || opts.method === "POST") {
+      return { output: { task_id: "task-1", task_status: "PENDING" } };
+    }
+    return {
+      output: { task_id: "task-1", task_status: "SUCCEEDED", results: [] },
+      request_id: "r1",
+    };
+  });
+
+  await speechRecognize(
+    env,
+    {
+      url: "https://example.com/a.wav",
+      model: "qwen-audio-3.0-asr-flash-filetrans",
+      vocabulary: { 鲸落: 4 },
+      context: "鲸落 深海勇士",
+      "poll-interval": 0,
+    },
+    makeCtx(),
+  );
+
+  expect(captured[0]?.body).toMatchObject({
+    input: {
+      file_urls: ["https://example.com/a.wav"],
+      context: [{ role: "user", content: [{ type: "input_text", text: "鲸落 深海勇士" }] }],
+    },
+    parameters: { vocabulary: { 鲸落: 4 } },
+  });
+});
+
+test("pipeline speechRecognize rejects vocabulary on qwen3 sync flash", async () => {
+  const { env, captured } = makeEnv();
+  await expect(
+    speechRecognize(
+      env,
+      {
+        url: "https://example.com/a.wav",
+        model: "qwen3-asr-flash",
+        vocabulary: { 奋斗者: 4 },
+      },
+      makeCtx(),
+    ),
+  ).rejects.toBeInstanceOf(PipelineError);
+  expect(captured).toHaveLength(0);
+});
+
 test("pipeline speechRecognize maps qwen3-filetrans language to parameters.language", async () => {
   const { env, captured } = makeEnv(async (opts) => {
     if (opts.async || opts.method === "POST") {

packages/runtime/tests/watermark-pipeline.test.ts

@@ -0,0 +1,129 @@
+import { expect, test } from "vite-plus/test";
+import type { Client } from "bailian-cli-core";
+import { applyProfileWatermarkToStepInput, type PipelineEnv } from "../src/pipeline/bl-config.ts";
+import { imageEdit, imageGenerate, videoGenerate } from "../src/pipeline/steps/bl-api.ts";
+import type { StepContext } from "../src/pipeline/types.ts";
+
+type CapturedRequest = {
+  path?: string;
+  method?: string;
+  body?: {
+    parameters?: { watermark?: boolean };
+  };
+  async?: boolean;
+};
+
+function makeEnv(watermark: boolean): {
+  env: PipelineEnv;
+  captured: CapturedRequest[];
+} {
+  const captured: CapturedRequest[] = [];
+  const client = {
+    uploadFile: async (source: string) => source,
+    requestJson: async (opts: CapturedRequest) => {
+      captured.push(opts);
+      if (opts.async) {
+        return { output: { task_id: "task-wm", task_status: "PENDING" } };
+      }
+      // Sync image response shape (qwen-image / wan2.7-image).
+      return {
+        request_id: "req-wm",
+        output: {
+          choices: [{ message: { content: [{ image: "https://example.com/out.png" }] } }],
+        },
+      };
+    },
+  } as unknown as Client;
+
+  return {
+    env: {
+      client,
+      settings: {
+        quiet: true,
+        output: "json",
+        outputExplicit: true,
+        timeout: 300,
+        watermark,
+        verbose: false,
+        dryRun: false,
+        telemetry: false,
+      } as PipelineEnv["settings"],
+    },
+    captured,
+  };
+}
+
+function makeCtx(): StepContext {
+  return { dryRun: false, signal: new AbortController().signal, timeoutSeconds: 1 };
+}
+
+test("pipeline imageGenerate inherits Profile watermark=false when step omits it", async () => {
+  const { env, captured } = makeEnv(false);
+  await imageGenerate(env, { prompt: "A cat" }, makeCtx());
+  expect(captured[0]?.body?.parameters?.watermark).toBe(false);
+});
+
+test("pipeline imageGenerate keeps step watermark over Profile", async () => {
+  const { env, captured } = makeEnv(false);
+  await imageGenerate(env, { prompt: "A cat", watermark: true }, makeCtx());
+  expect(captured[0]?.body?.parameters?.watermark).toBe(true);
+});
+
+test("pipeline imageEdit inherits Profile watermark=false when step omits it", async () => {
+  const { env, captured } = makeEnv(false);
+  await imageEdit(env, { prompt: "Blue sky", image: "https://example.com/in.png" }, makeCtx());
+  expect(captured[0]?.body?.parameters?.watermark).toBe(false);
+});
+
+test("pipeline videoGenerate inherits Profile watermark=false when step omits it", async () => {
+  const { env, captured } = makeEnv(false);
+  // First call submits async task; pollTaskWithOptions will keep polling — mock SUCCEEDED quickly.
+  const client = env.client as unknown as {
+    requestJson: (opts: CapturedRequest) => Promise<unknown>;
+  };
+  let calls = 0;
+  client.requestJson = async (opts: CapturedRequest) => {
+    captured.push(opts);
+    calls += 1;
+    if (calls === 1) {
+      return { output: { task_id: "task-wm", task_status: "PENDING" } };
+    }
+    return {
+      output: {
+        task_id: "task-wm",
+        task_status: "SUCCEEDED",
+        video_url: "https://example.com/out.mp4",
+      },
+    };
+  };
+
+  await videoGenerate(
+    env,
+    { prompt: "A cat walks", "poll-interval": 0 },
+    { ...makeCtx(), timeoutSeconds: 5 },
+  );
+  expect(captured[0]?.body?.parameters?.watermark).toBe(false);
+});
+
+test("pipeline defaults watermark to true when Profile leaves the compliance default", async () => {
+  const { env, captured } = makeEnv(true);
+  await imageGenerate(env, { prompt: "A cat" }, makeCtx());
+  expect(captured[0]?.body?.parameters?.watermark).toBe(true);
+});
+
+test("applyProfileWatermarkToStepInput fills omitted watermark from Settings", () => {
+  const settings = { watermark: false } as PipelineEnv["settings"];
+  expect(
+    applyProfileWatermarkToStepInput("image/generate", { prompt: "A cat" }, settings).watermark,
+  ).toBe(false);
+  expect(
+    applyProfileWatermarkToStepInput(
+      "image/generate",
+      { prompt: "A cat", watermark: true },
+      settings,
+    ).watermark,
+  ).toBe(true);
+  expect(
+    applyProfileWatermarkToStepInput("text/chat", { message: "hi" }, settings).watermark,
+  ).toBeUndefined();
+});

pnpm-lock.yaml

@@ -189,8 +189,8 @@ importers:
   packages/commands:
     dependencies:
       '@openagentpack/sdk':
-        specifier: 0.5.0
-        version: 0.5.0
+        specifier: 0.7.1
+        version: 0.7.1
       bailian-cli-core:
         specifier: workspace:*
         version: link:../core
@@ -1124,8 +1124,8 @@ packages:
       '@emnapi/core': ^1.7.1
       '@emnapi/runtime': ^1.7.1
 
-  '@openagentpack/sdk@0.5.0':
-    resolution: {integrity: sha512-9mNMvPWuoiK5NQVFN5i2qAhUAFSmw9jO8xrMDXXfa6eSu0XlEAPZzR6DDg4HdhW+ex60KgA+fcNA0h0LRF552A==}
+  '@openagentpack/sdk@0.7.1':
+    resolution: {integrity: sha512-OENks8UdfWFanvhJIVbqyUX1+XGavHnrxDiFx4SoJoGRXi45cOgyYtWWbG5VdDI2I5Dpo9eGNaw/JtDnkdxUcg==}
     engines: {node: '>=18.17.0'}
 
   '@oxc-project/runtime@0.129.0':
@@ -4287,7 +4287,7 @@ snapshots:
       '@tybys/wasm-util': 0.10.1
     optional: true
 
-  '@openagentpack/sdk@0.5.0':
+  '@openagentpack/sdk@0.7.1':
     dependencies:
       jszip: 3.10.1
       yaml: 2.9.0

skills/bailian-cli/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: bailian-cli
 metadata:
-  version: "1.20.0"
+  version: "1.23.0"
   requires:
     bins: ["bl"]
 description: >-

skills/bailian-cli/reference/config.md

@@ -88,10 +88,10 @@ bl config list --output json
 
 #### Flags
 
-| Flag              | Type   | Required | Description                                                                                                                                                                                  |
-| ----------------- | ------ | -------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `--key <key>`     | string | yes      | Config key (language, base*url, output, output_dir, timeout, api_key, api_key_capabilities, access_token, access_key_id, access_key_secret, security_token, default*\*\_model, workspace_id) |
-| `--value <value>` | string | yes      | Value to set                                                                                                                                                                                 |
+| Flag              | Type   | Required | Description                                                                                                                                                                                             |
+| ----------------- | ------ | -------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `--key <key>`     | string | yes      | Config key (language, base*url, output, output_dir, timeout, watermark, api_key, api_key_capabilities, access_token, access_key_id, access_key_secret, security_token, default*\*\_model, workspace_id) |
+| `--value <value>` | string | yes      | Value to set                                                                                                                                                                                            |
 
 #### Examples
 
@@ -107,6 +107,10 @@ bl config set --key output --value json
 bl config set --key timeout --value 600
 ```
 
+```bash
+bl config set --key watermark --value false
+```
+
 ```bash
 bl config set --key base_url --value https://dashscope.aliyuncs.com
 ```

skills/bailian-finetune/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: bailian-finetune
 metadata:
-  version: "1.20.0"
+  version: "1.23.0"
   requires:
     bins: ["bl"]
 description: >-

skills/bailian-gen/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: bailian-gen
 metadata:
-  version: "1.20.0"
+  version: "1.23.0"
   requires:
     bins: ["bl"]
 description: >-
@@ -45,8 +45,27 @@ Unless the user explicitly specifies a model, omit `--model` and let the CLI use
 
 For ASR model selection, keep `fun-asr` (or other `*-filetrans`) for long recordings, repeated files, speaker diarization, or asynchronous task IDs. For one local or remote audio file up to about five minutes when the user asks for low-latency Flash models, use `--model fun-asr-flash-2026-06-15`, `--model qwen-audio-3.0-asr-flash`, or `--model qwen3-asr-flash`. Flash recognition is synchronous and accepts exactly one file per call.
 
+To improve ASR accuracy with domain terms:
+
+- Prefer instant `--vocabulary` / `--context` on `bl speech recognize` when the model is Qwen-Audio-3.0-ASR-Flash series (and Fun-ASR-Flash for `--context` only) — no pre-built vocabulary needed. Start weights at 4 (do not default everything to 5). `--context` must list the target words themselves; a topic description alone has little effect.
+- Use `bl speech vocabulary create` + `--vocabulary-id` for Fun-ASR / Paraformer, or whenever the same hot words must be reused across requests. The vocabulary `--model` must exactly match recognize `--model` (otherwise the vocabulary is silently ignored). Each account may have at most 10 vocabularies; delete unused ones.
+
 Flags, usage, and examples: see [`reference/`](reference/index.md) or `bl <command> --help` — do not guess flags.
 
+## Watermark configuration
+
+Image generation and editing, video generation and editing, and reference-to-video enable watermarks by default. Change the default for the active Profile with:
+
+```bash
+bl config set --key watermark --value false
+```
+
+Set it to `true` to enable watermarks again. To update a named Profile without switching Profiles, add `--config <name>`:
+
+```bash
+bl config set --config media --key watermark --value false
+```
+
 ## Local files (mandatory)
 
 Any command that accepts a **file URL** also accepts a **local path**; the CLI uploads to DashScope temporary storage (`oss://`, 48h) automatically. If the user gives a local file, pass the path directly — never ask them to upload or host a URL first.
@@ -68,6 +87,11 @@ bl vision describe --image ./photo.jpg --prompt "图里有什么？"
 bl vision describe --video ./clip.mp4 --prompt "总结视频内容"
 bl omni --message "Describe the video content" --video ./demo.mp4 --text-only
 bl speech synthesize --text "Hello, welcome to Bailian" --out hello.mp3
+bl speech recognize --url ./meeting.wav --model qwen-audio-3.0-asr-flash-filetrans \
+  --vocabulary '{"奋斗者":4}' --context "奋斗者号"
+VOCAB=$(bl speech vocabulary create --model fun-asr --prefix demo --words '{"奋斗者":4}' --quiet)
+bl speech recognize --url ./meeting.wav --model fun-asr --vocabulary-id "$VOCAB"
+bl speech vocabulary delete --id "$VOCAB" --yes
 ```
 
 ## Output language

skills/bailian-gen/reference/image.md

@@ -36,7 +36,7 @@ Index: [index.md](index.md)
 | `--negative-prompt <text>`  | string  | no       | Negative prompt to exclude unwanted content                                                        |
 | `--function <name>`         | string  | no       | wanx\*-imageedit function (default: description_edit). Examples: stylization_all, description_edit |
 | `--prompt-extend <bool>`    | boolean | no       | Enable prompt extend (true/false). Omit flag to use CLI default (true).                            |
-| `--watermark <bool>`        | boolean | no       | Enable watermark (true/false). Omit flag to use CLI default (true).                                |
+| `--watermark <bool>`        | boolean | no       | Enable watermark (true/false). Omit flag to use the Profile setting or CLI default (true).         |
 | `--out-dir <dir>`           | string  | no       | Download images to directory                                                                       |
 | `--out-prefix <prefix>`     | string  | no       | Filename prefix (default: edited)                                                                  |
 | `--async`                   | switch  | no       | Return async task id without waiting                                                               |
@@ -99,7 +99,7 @@ bl image edit --image ./photo.png --prompt "Replace the background with a beach"
 | `--seed <n>`                | number  | no       | Random seed for reproducible generation                                                                                  |
 | `--negative-prompt <text>`  | string  | no       | Negative prompt to exclude unwanted content                                                                              |
 | `--prompt-extend <bool>`    | boolean | no       | Enable prompt extend (true/false). Omit flag: true for qwen-image sync; parameter omitted on async models (API default). |
-| `--watermark <bool>`        | boolean | no       | Enable watermark (true/false). Omit flag to use CLI default (true).                                                      |
+| `--watermark <bool>`        | boolean | no       | Enable watermark (true/false). Omit flag to use the Profile setting or CLI default (true).                               |
 | `--async`                   | switch  | no       | Return async task id without waiting                                                                                     |
 | `--concurrent <n>`          | number  | no       | Run N parallel requests (default: 1)                                                                                     |
 | `--out-dir <dir>`           | string  | no       | Download images to directory                                                                                             |

skills/bailian-gen/reference/index.md

@@ -9,29 +9,34 @@ Use this index for the skill-scoped quick index and global flags.
 
 ## Quick index
 
-| Command                | Authentication | Description                                                                                                          | Detail                 |
-| ---------------------- | -------------- | -------------------------------------------------------------------------------------------------------------------- | ---------------------- |
-| `bl image edit`        | API Key        | Edit an existing image with text instructions (Qwen-Image / Wan 2.7)                                                 | [image.md](image.md)   |
-| `bl image generate`    | API Key        | Generate images (Qwen-Image / wan2.x)                                                                                | [image.md](image.md)   |
-| `bl omni`              | API Key        | Multimodal chat with text + audio output (Qwen-Omni)                                                                 | [omni.md](omni.md)     |
-| `bl speech recognize`  | API Key        | Recognize speech from audio files (FunAudio-ASR / Qwen-ASR Flash)                                                    | [speech.md](speech.md) |
-| `bl speech synthesize` | API Key        | Synthesize speech from text                                                                                          | [speech.md](speech.md) |
-| `bl video download`    | API Key        | Download a completed video by task ID                                                                                | [video.md](video.md)   |
-| `bl video edit`        | API Key        | Edit a video with happyhorse-1.0-video-edit (style transfer, object replacement, etc.)                               | [video.md](video.md)   |
-| `bl video generate`    | API Key        | Generate a video from text or image (wan3.0-video / wan2.6-t2v / happyhorse-1.1-i2v)                                 | [video.md](video.md)   |
-| `bl video ref`         | API Key        | Reference-to-video generation (wan3.0-video / happyhorse-1.1-r2v / wan2.6-r2v): multi-subject, multi-shot with voice | [video.md](video.md)   |
-| `bl video task get`    | API Key        | Query async task status                                                                                              | [video.md](video.md)   |
-| `bl vision describe`   | API Key        | Describe an image or video using Qwen-VL                                                                             | [vision.md](vision.md) |
+| Command                       | Authentication | Description                                                                                                          | Detail                 |
+| ----------------------------- | -------------- | -------------------------------------------------------------------------------------------------------------------- | ---------------------- |
+| `bl image edit`               | API Key        | Edit an existing image with text instructions (Qwen-Image / Wan 2.7)                                                 | [image.md](image.md)   |
+| `bl image generate`           | API Key        | Generate images (Qwen-Image / wan2.x)                                                                                | [image.md](image.md)   |
+| `bl omni`                     | API Key        | Multimodal chat with text + audio output (Qwen-Omni)                                                                 | [omni.md](omni.md)     |
+| `bl speech recognize`         | API Key        | Recognize speech from audio files (FunAudio-ASR / Qwen-ASR Flash)                                                    | [speech.md](speech.md) |
+| `bl speech synthesize`        | API Key        | Synthesize speech from text                                                                                          | [speech.md](speech.md) |
+| `bl speech vocabulary create` | API Key        | Create a precompiled hot-word vocabulary for ASR                                                                     | [speech.md](speech.md) |
+| `bl speech vocabulary delete` | API Key        | Delete a precompiled hot-word vocabulary                                                                             | [speech.md](speech.md) |
+| `bl speech vocabulary get`    | API Key        | Get details of a precompiled hot-word vocabulary                                                                     | [speech.md](speech.md) |
+| `bl speech vocabulary list`   | API Key        | List precompiled hot-word vocabularies                                                                               | [speech.md](speech.md) |
+| `bl speech vocabulary update` | API Key        | Replace the contents of a precompiled hot-word vocabulary                                                            | [speech.md](speech.md) |
+| `bl video download`           | API Key        | Download a completed video by task ID                                                                                | [video.md](video.md)   |
+| `bl video edit`               | API Key        | Edit a video with happyhorse-1.0-video-edit (style transfer, object replacement, etc.)                               | [video.md](video.md)   |
+| `bl video generate`           | API Key        | Generate a video from text or image (wan3.0-video / wan2.6-t2v / happyhorse-1.1-i2v)                                 | [video.md](video.md)   |
+| `bl video ref`                | API Key        | Reference-to-video generation (wan3.0-video / happyhorse-1.1-r2v / wan2.6-r2v): multi-subject, multi-shot with voice | [video.md](video.md)   |
+| `bl video task get`           | API Key        | Query async task status                                                                                              | [video.md](video.md)   |
+| `bl vision describe`          | API Key        | Describe an image or video using Qwen-VL                                                                             | [vision.md](vision.md) |
 
 ## By group
 
-| Group    | Commands                                          | Reference              |
-| -------- | ------------------------------------------------- | ---------------------- |
-| `image`  | `edit`, `generate`                                | [image.md](image.md)   |
-| `omni`   | `(root)`                                          | [omni.md](omni.md)     |
-| `speech` | `recognize`, `synthesize`                         | [speech.md](speech.md) |
-| `video`  | `download`, `edit`, `generate`, `ref`, `task get` | [video.md](video.md)   |
-| `vision` | `describe`                                        | [vision.md](vision.md) |
+| Group    | Commands                                                                                                                      | Reference              |
+| -------- | ----------------------------------------------------------------------------------------------------------------------------- | ---------------------- |
+| `image`  | `edit`, `generate`                                                                                                            | [image.md](image.md)   |
+| `omni`   | `(root)`                                                                                                                      | [omni.md](omni.md)     |
+| `speech` | `recognize`, `synthesize`, `vocabulary create`, `vocabulary delete`, `vocabulary get`, `vocabulary list`, `vocabulary update` | [speech.md](speech.md) |
+| `video`  | `download`, `edit`, `generate`, `ref`, `task get`                                                                             | [video.md](video.md)   |
+| `vision` | `describe`                                                                                                                    | [vision.md](vision.md) |
 
 ## Global flags
 

skills/bailian-gen/reference/speech.md

@@ -7,10 +7,15 @@ Index: [index.md](index.md)
 
 ## Commands in this group
 
-| Command                | Authentication | Description                                                       |
-| ---------------------- | -------------- | ----------------------------------------------------------------- |
-| `bl speech recognize`  | API Key        | Recognize speech from audio files (FunAudio-ASR / Qwen-ASR Flash) |
-| `bl speech synthesize` | API Key        | Synthesize speech from text                                       |
+| Command                       | Authentication | Description                                                       |
+| ----------------------------- | -------------- | ----------------------------------------------------------------- |
+| `bl speech recognize`         | API Key        | Recognize speech from audio files (FunAudio-ASR / Qwen-ASR Flash) |
+| `bl speech synthesize`        | API Key        | Synthesize speech from text                                       |
+| `bl speech vocabulary create` | API Key        | Create a precompiled hot-word vocabulary for ASR                  |
+| `bl speech vocabulary delete` | API Key        | Delete a precompiled hot-word vocabulary                          |
+| `bl speech vocabulary get`    | API Key        | Get details of a precompiled hot-word vocabulary                  |
+| `bl speech vocabulary list`   | API Key        | List precompiled hot-word vocabularies                            |
+| `bl speech vocabulary update` | API Key        | Replace the contents of a precompiled hot-word vocabulary         |
 
 ## Command details
 
@@ -25,20 +30,22 @@ Index: [index.md](index.md)
 
 #### Flags
 
-| Flag                        | Type   | Required | Description                                                                                                                                                                         |
-| --------------------------- | ------ | -------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `--url <url>`               | array  | yes      | Audio file URL or local file path (repeatable, max 100)                                                                                                                             |
-| `--model <model>`           | string | no       | Model ID (default: configured Profile ASR model, otherwise fun-asr). Async: fun-asr / _-filetrans / paraformer-_; sync: qwen3-asr-flash* / fun-asr-flash* / qwen-audio-\*-asr-flash |
-| `--language <lang>`         | string | no       | Language hint (e.g. zh, en, ja). Classic async/input-audio: language_hints; qwen3-filetrans: language; qwen3 sync: asr_options.language                                             |
-| `--diarization`             | switch | no       | Enable automatic speaker diarization                                                                                                                                                |
-| `--speaker-count <n>`       | number | no       | Expected number of speakers (requires --diarization)                                                                                                                                |
-| `--vocabulary-id <id>`      | string | no       | Hot-word vocabulary ID for improved accuracy                                                                                                                                        |
-| `--channel-id <n>`          | number | no       | Audio channel ID (default: 0)                                                                                                                                                       |
-| `--out <path>`              | string | no       | Save full transcription result to JSON file                                                                                                                                         |
-| `--async`                   | switch | no       | Return async task id without waiting                                                                                                                                                |
-| `--poll-interval <seconds>` | number | no       | Polling interval in seconds (default: 2)                                                                                                                                            |
-| `--api-key <key>`           | string | no       | API key                                                                                                                                                                             |
-| `--base-url <url>`          | string | no       | API base URL                                                                                                                                                                        |
+| Flag                        | Type   | Required | Description                                                                                                                                                                                                                                          |
+| --------------------------- | ------ | -------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `--url <url>`               | array  | yes      | Audio file URL or local file path (repeatable, max 100)                                                                                                                                                                                              |
+| `--model <model>`           | string | no       | Model ID (default: configured Profile ASR model, otherwise fun-asr). Async: fun-asr / _-filetrans / paraformer-_; sync: qwen3-asr-flash* / fun-asr-flash* / qwen-audio-\*-asr-flash                                                                  |
+| `--language <lang>`         | string | no       | Language hint (e.g. zh, en, ja). Classic async/input-audio: language_hints; qwen3-filetrans: language; qwen3 sync: asr_options.language                                                                                                              |
+| `--diarization`             | switch | no       | Enable automatic speaker diarization                                                                                                                                                                                                                 |
+| `--speaker-count <n>`       | number | no       | Expected number of speakers (requires --diarization)                                                                                                                                                                                                 |
+| `--vocabulary-id <id>`      | string | no       | Pre-built hot-word vocabulary ID (create it via `speech vocabulary create`). Its target_model must exactly match --model, otherwise it is silently ignored. Wider model support than --vocabulary, including Fun-ASR and Paraformer                  |
+| `--vocabulary <json>`       | string | no       | Instant hot words as JSON object of word→weight, e.g. '{"Fendouzhe":4}'. Weight 1-5 (4 recommended; higher values can hurt other words), 50 for super hot word. No pre-built vocabulary needed. Takes effect only on Qwen-Audio-3.0-ASR-Flash models |
+| `--context <text>`          | string | no       | Context enhancement word list to improve accuracy on proper nouns; must contain the target words themselves (a topic description alone has little effect); max 400 chars. Takes effect only on Qwen-Audio-3.0-ASR-Flash and Fun-ASR-Flash models     |
+| `--channel-id <n>`          | number | no       | Audio channel ID (default: 0)                                                                                                                                                                                                                        |
+| `--out <path>`              | string | no       | Save full transcription result to JSON file                                                                                                                                                                                                          |
+| `--async`                   | switch | no       | Return async task id without waiting                                                                                                                                                                                                                 |
+| `--poll-interval <seconds>` | number | no       | Polling interval in seconds (default: 2)                                                                                                                                                                                                             |
+| `--api-key <key>`           | string | no       | API key                                                                                                                                                                                                                                              |
+| `--base-url <url>`          | string | no       | API base URL                                                                                                                                                                                                                                         |
 
 #### Examples
 
@@ -62,6 +69,14 @@ bl speech recognize --url https://example.com/audio.mp3 --language zh
 bl speech recognize --url https://example.com/audio.mp3 --vocabulary-id vocab-abc123
 ```
 
+```bash
+bl speech recognize --url https://example.com/audio.mp3 --model qwen-audio-3.0-asr-flash-filetrans --vocabulary '{"奋斗者":4,"鲸落":4}'
+```
+
+```bash
+bl speech recognize --url https://example.com/audio.mp3 --model qwen-audio-3.0-asr-flash-filetrans --context "奋斗者号 鲸落 深海勇士"
+```
+
 ```bash
 bl speech recognize --url https://example.com/audio.mp3 --out result.json
 ```
@@ -148,3 +163,198 @@ bl speech synthesize --text "Hello" --voice <voice_id> --stream | afplay -
 ```bash
 bl speech synthesize --text "Hello" --voice <voice_id> --stream | ffplay -nodisp -autoexit -f s16le -ar 24000 -ac 1 -
 ```
+
+### `bl speech vocabulary create`
+
+| Field              | Value                                                                                                           |
+| ------------------ | --------------------------------------------------------------------------------------------------------------- |
+| **Name**           | `speech vocabulary create`                                                                                      |
+| **Description**    | Create a precompiled hot-word vocabulary for ASR                                                                |
+| **Authentication** | API Key                                                                                                         |
+| **Usage**          | `bl speech vocabulary create --model <model> --prefix <prefix> (--words <json> \| --words-file <path>) [flags]` |
+
+#### Flags
+
+| Flag                  | Type   | Required | Description                                                                                                                                                                                                                                                                    |
+| --------------------- | ------ | -------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| `--model <model>`     | string | yes      | ASR model this vocabulary is built for (required). Must exactly match the --model passed to `speech recognize` later, otherwise the vocabulary is silently ignored                                                                                                             |
+| `--prefix <prefix>`   | string | yes      | Custom vocabulary prefix (required). Digits and lowercase letters only, max 10 chars                                                                                                                                                                                           |
+| `--words <json>`      | string | no       | Hot words as JSON object of word→weight, e.g. '{"Fendouzhe":4}'; or the API entry array for per-entry lang. Weight 1-5 (4 recommended); when the vocabulary target_model is a Qwen-Audio-3.0-ASR-Flash series model, 50 is also allowed as super hot word. Or use --words-file |
+| `--words-file <path>` | string | no       | JSON file with the hot words (use - for stdin)                                                                                                                                                                                                                                 |
+| `--lang <code>`       | string | no       | Language code applied to every hot word when using object form (optional; ignored for array form). Paraformer: zh/en/ja/yue/ko/de/fr/ru; Fun-ASR: zh/en/ja                                                                                                                     |
+| `--api-key <key>`     | string | no       | API key                                                                                                                                                                                                                                                                        |
+| `--base-url <url>`    | string | no       | API base URL                                                                                                                                                                                                                                                                   |
+
+#### Notes
+
+- The --model must exactly match the --model used later with `speech recognize --vocabulary-id`; a mismatch causes silent failure with no error.
+- Each account may have at most 10 vocabularies; updates must be at least 5 minutes apart. See improve-asr-accuracy for full limits.
+- Hot-word vocabularies are not supported in Singapore sub-workspaces; the server error is passed through as-is.
+- Weight 1-5 (4 recommended); when the vocabulary target_model is a Qwen-Audio-3.0-ASR-Flash series model, 50 is also allowed as super hot word.
+
+#### Examples
+
+```bash
+bl speech vocabulary create --model fun-asr --prefix demo --words '{"Fendouzhe":4,"Jingluo":4}'
+```
+
+```bash
+bl speech vocabulary create --model paraformer-v2 --prefix demo --words '[{"text":"Fendouzhe","weight":4,"lang":"zh"}]'
+```
+
+```bash
+bl speech vocabulary create --model fun-asr --prefix demo --words '{"Fendouzhe":4}' --lang zh
+```
+
+```bash
+bl speech vocabulary create --model fun-asr --prefix demo --words-file ./hotwords.json
+```
+
+```bash
+bl speech vocabulary create --model fun-asr --prefix demo --words '{"Fendouzhe":4}' --quiet
+```
+
+### `bl speech vocabulary delete`
+
+| Field              | Value                                                                            |
+| ------------------ | -------------------------------------------------------------------------------- |
+| **Name**           | `speech vocabulary delete`                                                       |
+| **Description**    | Delete a precompiled hot-word vocabulary                                         |
+| **Authentication** | API Key                                                                          |
+| **Usage**          | `bl speech vocabulary delete --id <id>`                                          |
+| **Risk**           | `high`                                                                           |
+| **Risk message**   | This permanently deletes the specified hot-word vocabulary and cannot be undone. |
+
+> **Agent safety:** Never add `--yes` automatically. On `type="requires_confirmation"`, stop and ask for explicit user confirmation of the same action and scope.
+
+#### Flags
+
+| Flag               | Type   | Required | Description                       |
+| ------------------ | ------ | -------- | --------------------------------- |
+| `--id <id>`        | string | yes      | Hot-word vocabulary ID (required) |
+| `--yes`            | switch | no       | Confirm this high-risk operation  |
+| `--api-key <key>`  | string | no       | API key                           |
+| `--base-url <url>` | string | no       | API base URL                      |
+
+#### Examples
+
+```bash
+bl speech vocabulary delete --id vocab-demo-xxx --dry-run
+```
+
+```bash
+# Only after explicit user confirmation:
+bl speech vocabulary delete --id vocab-demo-xxx --yes
+```
+
+### `bl speech vocabulary get`
+
+| Field              | Value                                            |
+| ------------------ | ------------------------------------------------ |
+| **Name**           | `speech vocabulary get`                          |
+| **Description**    | Get details of a precompiled hot-word vocabulary |
+| **Authentication** | API Key                                          |
+| **Usage**          | `bl speech vocabulary get --id <id>`             |
+
+#### Flags
+
+| Flag               | Type   | Required | Description                       |
+| ------------------ | ------ | -------- | --------------------------------- |
+| `--id <id>`        | string | yes      | Hot-word vocabulary ID (required) |
+| `--api-key <key>`  | string | no       | API key                           |
+| `--base-url <url>` | string | no       | API base URL                      |
+
+#### Notes
+
+- Use this command to confirm target_model before calling `speech recognize --vocabulary-id`; a model mismatch causes silent failure.
+
+#### Examples
+
+```bash
+bl speech vocabulary get --id vocab-demo-xxx
+```
+
+```bash
+bl speech vocabulary get --id vocab-demo-xxx --quiet
+```
+
+### `bl speech vocabulary list`
+
+| Field              | Value                                                                          |
+| ------------------ | ------------------------------------------------------------------------------ |
+| **Name**           | `speech vocabulary list`                                                       |
+| **Description**    | List precompiled hot-word vocabularies                                         |
+| **Authentication** | API Key                                                                        |
+| **Usage**          | `bl speech vocabulary list [--prefix <prefix>] [--page <n>] [--page-size <n>]` |
+
+#### Flags
+
+| Flag                | Type   | Required | Description                                                                       |
+| ------------------- | ------ | -------- | --------------------------------------------------------------------------------- |
+| `--prefix <prefix>` | string | no       | Filter by vocabulary prefix                                                       |
+| `--page <n>`        | number | no       | Page number, 1-based (default: 1). Mapped to API page_index (0-based) as page - 1 |
+| `--page-size <n>`   | number | no       | Results per page (default: 10)                                                    |
+| `--api-key <key>`   | string | no       | API key                                                                           |
+| `--base-url <url>`  | string | no       | API base URL                                                                      |
+
+#### Notes
+
+- List responses do not include target_model; use `speech vocabulary get` to inspect the model a vocabulary was built for.
+- Vocabularies with status UNDEPLOYED are silently ignored by ASR.
+
+#### Examples
+
+```bash
+bl speech vocabulary list
+```
+
+```bash
+bl speech vocabulary list --prefix demo
+```
+
+```bash
+bl speech vocabulary list --page 2 --page-size 20
+```
+
+### `bl speech vocabulary update`
+
+| Field              | Value                                                                                                           |
+| ------------------ | --------------------------------------------------------------------------------------------------------------- |
+| **Name**           | `speech vocabulary update`                                                                                      |
+| **Description**    | Replace the contents of a precompiled hot-word vocabulary                                                       |
+| **Authentication** | API Key                                                                                                         |
+| **Usage**          | `bl speech vocabulary update --id <id> (--words <json> \| --words-file <path>) [flags]`                         |
+| **Risk**           | `high`                                                                                                          |
+| **Risk message**   | This fully replaces all hot words in the vocabulary. Entries not listed will be discarded and cannot be undone. |
+
+> **Agent safety:** Never add `--yes` automatically. On `type="requires_confirmation"`, stop and ask for explicit user confirmation of the same action and scope.
+
+#### Flags
+
+| Flag                  | Type   | Required | Description                                                                                                                                                                                                                                                                    |
+| --------------------- | ------ | -------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| `--id <id>`           | string | yes      | Hot-word vocabulary ID (required)                                                                                                                                                                                                                                              |
+| `--words <json>`      | string | no       | Hot words as JSON object of word→weight, e.g. '{"Fendouzhe":4}'; or the API entry array for per-entry lang. Weight 1-5 (4 recommended); when the vocabulary target_model is a Qwen-Audio-3.0-ASR-Flash series model, 50 is also allowed as super hot word. Or use --words-file |
+| `--words-file <path>` | string | no       | JSON file with the hot words (use - for stdin)                                                                                                                                                                                                                                 |
+| `--lang <code>`       | string | no       | Language code applied to every hot word when using object form (optional; ignored for array form). Paraformer: zh/en/ja/yue/ko/de/fr/ru; Fun-ASR: zh/en/ja                                                                                                                     |
+| `--yes`               | switch | no       | Confirm this high-risk operation                                                                                                                                                                                                                                               |
+| `--api-key <key>`     | string | no       | API key                                                                                                                                                                                                                                                                        |
+| `--base-url <url>`    | string | no       | API base URL                                                                                                                                                                                                                                                                   |
+
+#### Notes
+
+- update is a full replace, not an append. Prefer --dry-run first to preview the complete vocabulary that will be written.
+- Each account may have at most 10 vocabularies; updates must be at least 5 minutes apart. See improve-asr-accuracy for full limits.
+- Hot-word vocabularies are not supported in Singapore sub-workspaces; the server error is passed through as-is.
+- Weight 1-5 (4 recommended); when the vocabulary target_model is a Qwen-Audio-3.0-ASR-Flash series model, 50 is also allowed as super hot word.
+
+#### Examples
+
+```bash
+bl speech vocabulary update --id vocab-demo-xxx --words '{"Fendouzhe":4}' --dry-run
+```
+
+```bash
+# Only after explicit user confirmation:
+bl speech vocabulary update --id vocab-demo-xxx --words '{"Fendouzhe":4,"Jingluo":4}' --yes
+```

skills/bailian-gen/reference/video.md

@@ -56,26 +56,26 @@ bl video download --task-id 3b256896-xxxx --out video.mp4 --quiet
 
 #### Flags
 
-| Flag                             | Type    | Required | Description                                                                             |
-| -------------------------------- | ------- | -------- | --------------------------------------------------------------------------------------- |
-| `--model <model>`                | string  | no       | Model ID (default: happyhorse-1.0-video-edit)                                           |
-| `--video <url>`                  | string  | yes      | Input video URL or local file (mp4/mov, 2-10s)                                          |
-| `--prompt <text>`                | string  | no       | Edit instruction (e.g. "Convert the scene to a claymation style")                       |
-| `--ref-image <url>`              | string  | no       | Reference image URL (up to 4, comma-separated)                                          |
-| `--negative-prompt <text>`       | string  | no       | Negative prompt to exclude unwanted content                                             |
-| `--resolution <res>`             | string  | no       | Resolution: 720P or 1080P (default: 1080P)                                              |
-| `--ratio <ratio>`                | string  | no       | Aspect ratio (16:9, 9:16, 1:1, 4:3, 3:4)                                                |
-| `--duration <seconds>`           | number  | no       | Output video duration in seconds (2-10)                                                 |
-| `--audio-setting <auto\|origin>` | string  | no       | Audio: auto (default) or origin (keep original)                                         |
-| `--prompt-extend <bool>`         | boolean | no       | Enable prompt extend (true/false). Omit flag to omit the parameter (DashScope default). |
-| `--watermark <bool>`             | boolean | no       | Enable watermark (true/false). Omit flag to use CLI default (true).                     |
-| `--seed <n>`                     | number  | no       | Random seed for reproducible generation                                                 |
-| `--download <path>`              | string  | no       | Save video to file on completion                                                        |
-| `--async`                        | switch  | no       | Return async task id without waiting                                                    |
-| `--concurrent <n>`               | number  | no       | Run N parallel requests (default: 1)                                                    |
-| `--poll-interval <seconds>`      | number  | no       | Polling interval when waiting (default: 15)                                             |
-| `--api-key <key>`                | string  | no       | API key                                                                                 |
-| `--base-url <url>`               | string  | no       | API base URL                                                                            |
+| Flag                             | Type    | Required | Description                                                                                |
+| -------------------------------- | ------- | -------- | ------------------------------------------------------------------------------------------ |
+| `--model <model>`                | string  | no       | Model ID (default: happyhorse-1.0-video-edit)                                              |
+| `--video <url>`                  | string  | yes      | Input video URL or local file (mp4/mov, 2-10s)                                             |
+| `--prompt <text>`                | string  | no       | Edit instruction (e.g. "Convert the scene to a claymation style")                          |
+| `--ref-image <url>`              | string  | no       | Reference image URL (up to 4, comma-separated)                                             |
+| `--negative-prompt <text>`       | string  | no       | Negative prompt to exclude unwanted content                                                |
+| `--resolution <res>`             | string  | no       | Resolution: 720P or 1080P (default: 1080P)                                                 |
+| `--ratio <ratio>`                | string  | no       | Aspect ratio (16:9, 9:16, 1:1, 4:3, 3:4)                                                   |
+| `--duration <seconds>`           | number  | no       | Output video duration in seconds (2-10)                                                    |
+| `--audio-setting <auto\|origin>` | string  | no       | Audio: auto (default) or origin (keep original)                                            |
+| `--prompt-extend <bool>`         | boolean | no       | Enable prompt extend (true/false). Omit flag to omit the parameter (DashScope default).    |
+| `--watermark <bool>`             | boolean | no       | Enable watermark (true/false). Omit flag to use the Profile setting or CLI default (true). |
+| `--seed <n>`                     | number  | no       | Random seed for reproducible generation                                                    |
+| `--download <path>`              | string  | no       | Save video to file on completion                                                           |
+| `--async`                        | switch  | no       | Return async task id without waiting                                                       |
+| `--concurrent <n>`               | number  | no       | Run N parallel requests (default: 1)                                                       |
+| `--poll-interval <seconds>`      | number  | no       | Polling interval when waiting (default: 15)                                                |
+| `--api-key <key>`                | string  | no       | API key                                                                                    |
+| `--base-url <url>`               | string  | no       | API base URL                                                                               |
 
 #### Examples
 
@@ -117,7 +117,7 @@ bl video edit --video https://example.com/input.mp4 --prompt "Put clothes on the
 | `--ratio <ratio>`           | string  | no       | Aspect ratio (e.g. 16:9, 9:16, 1:1)                                                                                  |
 | `--duration <seconds>`      | number  | no       | Video duration in seconds (default: 5)                                                                               |
 | `--prompt-extend <bool>`    | boolean | no       | Enable prompt extend (true/false). Omit flag to omit the parameter (DashScope default).                              |
-| `--watermark <bool>`        | boolean | no       | Enable watermark (true/false). Omit flag to use CLI default (true).                                                  |
+| `--watermark <bool>`        | boolean | no       | Enable watermark (true/false). Omit flag to use the Profile setting or CLI default (true).                           |
 | `--seed <n>`                | number  | no       | Random seed for reproducible generation                                                                              |
 | `--download <path>`         | string  | no       | Save video to file on completion                                                                                     |
 | `--file <url-or-path>`      | string  | no       | Reference file URL or local path for file-to-video (wan3.0-video only; mutually exclusive with --image/--last-frame) |
@@ -172,7 +172,7 @@ bl video generate --prompt "A cat playing with a ball" --watermark false
 | `--ratio <ratio>`           | string  | no       | Aspect ratio (16:9, 9:16, 1:1)                                                                                                              |
 | `--duration <seconds>`      | number  | no       | Video duration in seconds (default: 5)                                                                                                      |
 | `--prompt-extend <bool>`    | boolean | no       | Enable prompt extend (true/false). Omit flag to omit the parameter (DashScope default).                                                     |
-| `--watermark <bool>`        | boolean | no       | Enable watermark (true/false). Omit flag to use CLI default (true).                                                                         |
+| `--watermark <bool>`        | boolean | no       | Enable watermark (true/false). Omit flag to use the Profile setting or CLI default (true).                                                  |
 | `--seed <n>`                | number  | no       | Random seed for reproducible generation                                                                                                     |
 | `--download <path>`         | string  | no       | Save video to file on completion                                                                                                            |
 | `--async`                   | switch  | no       | Return async task id without waiting                                                                                                        |

skills/bailian-managed-agent/SKILL.md

@@ -1,13 +1,14 @@
 ---
 name: bailian-managed-agent
 metadata:
-  version: "1.20.0"
+  version: "1.23.0"
   requires:
     bins: ["bl"]
 description: >-
   阿里云百炼托管 Agent 声明式基础设施与 API 命令入口：用户要创建agent、初始化 agents.yaml、校验或预览配置变更、
-  创建/更新/销毁托管 Agent 或 Deployment，或查询 Agent/Environment/Skill/Vault/Deployment、管理 Session/Event/File、
-  运行/暂停 Deployment 时使用 `bl managed-agent`。持久资源仍以 agents.yaml 为唯一事实源做 IaC；公开 API 能力按资源透出
+  创建/更新/销毁托管 Agent 或 Deployment、在 Workbench 编辑和调试目录项目、管理本地快照版本，或查询
+  Agent/Environment/Skill/Vault/Deployment、管理 Session/Event/File、运行/暂停 Deployment 时使用
+  `bl managed-agent`。持久资源仍以 agents.yaml 为唯一事实源做 IaC；公开 API 能力按资源透出
   list/get/search/versions/download、数据面和运行时动作命令。apply / destroy 与破坏性 API 命令受统一高风险确认闸门保护；
   务必先展示预览再让用户确认，禁止自动添加 `--yes`。
   反触发：调用已上线的百炼应用/智能体走 bailian-app-call 或 `bl app`；宿主 agent 自身的记忆、技能、
@@ -47,6 +48,31 @@ for explicit confirmation before re-running with `--yes`.
 6. Destroy   bl managed-agent destroy --yes # separate explicit confirmation
 ```
 
+## Directory projects, Workbench, and local versions
+
+| Intent                                  | Command                                                |
+| --------------------------------------- | ------------------------------------------------------ |
+| Create or convert a directory project   | `bl managed-agent project init`                        |
+| Validate and Build directory source     | `bl managed-agent project validate` / `build`          |
+| Publish the current immutable Build     | `bl managed-agent project publish --yes`               |
+| Launch project resource editing         | `bl managed-agent project workbench`                   |
+| Launch one Agent Session Preview        | `bl managed-agent playground --agent <id>`             |
+| Enable/disable project versions         | `bl managed-agent project version enable` / `disable`  |
+| Inspect local version state and history | `bl managed-agent project version status` / `list`     |
+| Preview or restore project source       | `bl managed-agent project version preview` / `restore` |
+
+- Bailian CLI and Workbench use the same `.openagentpack/versions/project` store and enable switch. Git is not required.
+- Directory projects always use Bailian. `project.json` does not declare a Provider; Build supplies the Bailian Provider configuration automatically.
+- Fresh `project init` includes complete Skill/File/Vault/Environment examples with bilingual README files under `agents/assistant/<resource-type>/_examples/`. They are not linked in `agent.json` and are excluded from Build discovery, Workbench declarations, and remote Publish. Copy a resource outside `_examples/` and configure its Agent reference to use it. Examples remain local versioned source; never put real secrets into them. / 新项目的四类资源示例默认不启用、不发布；请按 README 复制到 `_examples/` 外再配置引用，不要向示例写入真实密钥。
+- `project init`, `validate`, `build`, and version commands are local-only. Publish and Workbench resolve credentials from Bailian CLI flags, shell environment, or the active Profile; project initialization does not write credentials into the project directory.
+- Build is local-only. Publish never runs Build implicitly and consumes only a current `.openagentpack/build/agents.yaml` plus manifest.
+- Build moves literal Vault `secret_value` / `access_token` values from Agent-local or shared `vault.json` into project-root `.env`, replacing them with generated environment references. Existing references and `.env` entries are preserved; conflicts receive suffixed variable names. Preview/dry-run never write or print secrets. Publish and Workbench read the selected project's root `.env` as a fallback to inherited environment variables, even when invoked elsewhere. `.env` is private plaintext storage, excluded from local versions but not automatically ignored by Git; keep it backed up securely.
+- Build 会将 Agent 本地或共享 `vault.json` 中的明文密钥移入项目根目录 `.env`，再写回环境变量引用；保留已有引用和变量，重名时生成后缀。预览不写文件或输出密钥。`.env` 不进入版本快照，也不加密；请自行备份并加入 Git 忽略规则。
+- Agent-local File and Skill content supports Build-time association. A File may be copied directly into `agents/<agent>/files/`, or placed in `agents/<agent>/files/<id>/` when that directory contains exactly one content file; Build generates `file.json` and a `/mnt/<filename>` entry in `agent.json.files`. A directory under `agents/<agent>/skills/<id>/` containing `SKILL.md` generates `skill.json` and its `agent.json.skills` entry. Explicit JSON always wins; shared root resources remain explicit. Resources referenced by multiple Agents are promoted to the corresponding root shared directory during Build.
+- A successful Publish versions the canonical YAML and the complete project source tree, including Skill scripts/assets and binary files. Remote State is never versioned or restored.
+- `project version restore` restores source files to the working directory, invalidates Build, and does not move version history or remote State.
+- `managed-agent playground` remains the standalone `agents.yaml` Session Preview path; directory Workbench is only under `managed-agent project workbench`.
+
 ## Scoped single-resource create
 
 以下命令都先构造 `agents.yaml` 声明，再通过 SDK 的定向 Plan/Apply 创建远端资源，不绕过 State：

skills/bailian-managed-agent/reference/index.md

@@ -37,8 +37,20 @@ Use this index for the skill-scoped quick index and global flags.
 | `bl managed-agent file list`               | API Key        | List Managed Agent files                                                     | [managed-agent.md](managed-agent.md) |
 | `bl managed-agent file search`             | API Key        | Search Managed Agent files                                                   | [managed-agent.md](managed-agent.md) |
 | `bl managed-agent file upload`             | API Key        | Upload a Managed Agent file                                                  | [managed-agent.md](managed-agent.md) |
-| `bl managed-agent init`                    | No Auth        | Create a new agents.yaml template                                            | [managed-agent.md](managed-agent.md) |
+| `bl managed-agent init`                    | No Auth        | Create an agents.yaml template                                               | [managed-agent.md](managed-agent.md) |
 | `bl managed-agent plan`                    | API Key        | Show what changes would be applied to agent infrastructure                   | [managed-agent.md](managed-agent.md) |
+| `bl managed-agent playground`              | API Key        | Launch a Session Preview for an agents.yaml Agent                            | [managed-agent.md](managed-agent.md) |
+| `bl managed-agent project build`           | No Auth        | Organize directory source and generate the immutable Publish Build           | [managed-agent.md](managed-agent.md) |
+| `bl managed-agent project init`            | No Auth        | Create a directory project or convert the local agents.yaml                  | [managed-agent.md](managed-agent.md) |
+| `bl managed-agent project publish`         | API Key        | Publish the current directory-project Build and record a version             | [managed-agent.md](managed-agent.md) |
+| `bl managed-agent project validate`        | No Auth        | Validate a directory Agent project                                           | [managed-agent.md](managed-agent.md) |
+| `bl managed-agent project version disable` | No Auth        | Disable directory project versions                                           | [managed-agent.md](managed-agent.md) |
+| `bl managed-agent project version enable`  | No Auth        | Enable directory project versions                                            | [managed-agent.md](managed-agent.md) |
+| `bl managed-agent project version list`    | No Auth        | List directory project versions                                              | [managed-agent.md](managed-agent.md) |
+| `bl managed-agent project version preview` | No Auth        | Preview a directory project version                                          | [managed-agent.md](managed-agent.md) |
+| `bl managed-agent project version restore` | No Auth        | Restore a version to the project working directory                           | [managed-agent.md](managed-agent.md) |
+| `bl managed-agent project version status`  | No Auth        | Show directory project version status                                        | [managed-agent.md](managed-agent.md) |
+| `bl managed-agent project workbench`       | API Key        | Launch the directory project Workbench                                       | [managed-agent.md](managed-agent.md) |
 | `bl managed-agent session archive`         | API Key        | Archive a Managed Agent session                                              | [managed-agent.md](managed-agent.md) |
 | `bl managed-agent session create`          | API Key        | Create a new session for an agent                                            | [managed-agent.md](managed-agent.md) |
 | `bl managed-agent session debug`           | API Key        | Aggregate session diagnostics                                                | [managed-agent.md](managed-agent.md) |
@@ -74,9 +86,9 @@ Use this index for the skill-scoped quick index and global flags.
 
 ## By group
 
-| Group           | Commands                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                     | Reference                            |
-| --------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------ |
-| `managed-agent` | `agent create`, `agent get`, `agent list`, `agent search`, `agent versions`, `apply`, `deployment create`, `deployment get`, `deployment list`, `deployment pause`, `deployment run`, `deployment runs get`, `deployment runs list`, `deployment search`, `deployment unpause`, `destroy`, `environment create`, `environment get`, `environment list`, `environment search`, `file delete`, `file download`, `file get`, `file list`, `file search`, `file upload`, `init`, `plan`, `session archive`, `session create`, `session debug`, `session delete`, `session event list`, `session event send`, `session event stream`, `session events`, `session export`, `session get`, `session list`, `session run`, `session search`, `session send`, `session update`, `skill create`, `skill download`, `skill get`, `skill list`, `skill search`, `skill versions`, `skill-list`, `state import`, `state list`, `state rm`, `state show`, `validate`, `vault create`, `vault credential create`, `vault get`, `vault list`, `vault search` | [managed-agent.md](managed-agent.md) |
+| Group           | Commands                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                             | Reference                            |
+| --------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------ |
+| `managed-agent` | `agent create`, `agent get`, `agent list`, `agent search`, `agent versions`, `apply`, `deployment create`, `deployment get`, `deployment list`, `deployment pause`, `deployment run`, `deployment runs get`, `deployment runs list`, `deployment search`, `deployment unpause`, `destroy`, `environment create`, `environment get`, `environment list`, `environment search`, `file delete`, `file download`, `file get`, `file list`, `file search`, `file upload`, `init`, `plan`, `playground`, `project build`, `project init`, `project publish`, `project validate`, `project version disable`, `project version enable`, `project version list`, `project version preview`, `project version restore`, `project version status`, `project workbench`, `session archive`, `session create`, `session debug`, `session delete`, `session event list`, `session event send`, `session event stream`, `session events`, `session export`, `session get`, `session list`, `session run`, `session search`, `session send`, `session update`, `skill create`, `skill download`, `skill get`, `skill list`, `skill search`, `skill versions`, `skill-list`, `state import`, `state list`, `state rm`, `state show`, `validate`, `vault create`, `vault credential create`, `vault get`, `vault list`, `vault search` | [managed-agent.md](managed-agent.md) |
 
 ## Global flags
 

skills/bailian-managed-agent/reference/managed-agent.md

@@ -35,8 +35,20 @@ Index: [index.md](index.md)
 | `bl managed-agent file list`               | API Key        | List Managed Agent files                                                     |
 | `bl managed-agent file search`             | API Key        | Search Managed Agent files                                                   |
 | `bl managed-agent file upload`             | API Key        | Upload a Managed Agent file                                                  |
-| `bl managed-agent init`                    | No Auth        | Create a new agents.yaml template                                            |
+| `bl managed-agent init`                    | No Auth        | Create an agents.yaml template                                               |
 | `bl managed-agent plan`                    | API Key        | Show what changes would be applied to agent infrastructure                   |
+| `bl managed-agent playground`              | API Key        | Launch a Session Preview for an agents.yaml Agent                            |
+| `bl managed-agent project build`           | No Auth        | Organize directory source and generate the immutable Publish Build           |
+| `bl managed-agent project init`            | No Auth        | Create a directory project or convert the local agents.yaml                  |
+| `bl managed-agent project publish`         | API Key        | Publish the current directory-project Build and record a version             |
+| `bl managed-agent project validate`        | No Auth        | Validate a directory Agent project                                           |
+| `bl managed-agent project version disable` | No Auth        | Disable directory project versions                                           |
+| `bl managed-agent project version enable`  | No Auth        | Enable directory project versions                                            |
+| `bl managed-agent project version list`    | No Auth        | List directory project versions                                              |
+| `bl managed-agent project version preview` | No Auth        | Preview a directory project version                                          |
+| `bl managed-agent project version restore` | No Auth        | Restore a version to the project working directory                           |
+| `bl managed-agent project version status`  | No Auth        | Show directory project version status                                        |
+| `bl managed-agent project workbench`       | API Key        | Launch the directory project Workbench                                       |
 | `bl managed-agent session archive`         | API Key        | Archive a Managed Agent session                                              |
 | `bl managed-agent session create`          | API Key        | Create a new session for an agent                                            |
 | `bl managed-agent session debug`           | API Key        | Aggregate session diagnostics                                                |
@@ -1068,7 +1080,7 @@ bl managed-agent file upload --path ./report.pdf
 | Field              | Value                                                                   |
 | ------------------ | ----------------------------------------------------------------------- |
 | **Name**           | `managed-agent init`                                                    |
-| **Description**    | Create a new agents.yaml template                                       |
+| **Description**    | Create an agents.yaml template                                          |
 | **Authentication** | No Auth                                                                 |
 | **Usage**          | `bl managed-agent init [--agent-name <name>] [--file <path>] [--force]` |
 
@@ -1126,6 +1138,372 @@ bl managed-agent plan
 bl managed-agent plan --no-refresh
 ```
 
+### `bl managed-agent playground`
+
+| Field              | Value                                                                                 |
+| ------------------ | ------------------------------------------------------------------------------------- |
+| **Name**           | `managed-agent playground`                                                            |
+| **Description**    | Launch a Session Preview for an agents.yaml Agent                                     |
+| **Authentication** | API Key                                                                               |
+| **Usage**          | `bl managed-agent playground [--file <path>] [--agent <id>] [--port <n>] [--no-open]` |
+
+#### Flags
+
+| Flag               | Type   | Required | Description                                                           |
+| ------------------ | ------ | -------- | --------------------------------------------------------------------- |
+| `--file <path>`    | string | no       | Config file path (default: agents.yaml)                               |
+| `--port <n>`       | number | no       | Local port (default: 4848)                                            |
+| `--no-open`        | switch | no       | Do not open a browser automatically                                   |
+| `--agent <id>`     | string | no       | Agent to preview (required when the project declares multiple Agents) |
+| `--api-key <key>`  | string | no       | API key                                                               |
+| `--base-url <url>` | string | no       | API base URL                                                          |
+
+#### Notes
+
+- bl managed-agent supports the Bailian provider only; configurations containing other providers are rejected.
+- Bailian credentials come from bl's auth chain: --api-key > DASHSCOPE_API_KEY > `bl auth login` (active config profile).
+- Resolved credentials are injected into the SDK in-memory and cleared from the environment; they never persist in process env.
+- Session Preview requires Node.js 22+ and keeps using an agents.yaml source. Directory Workbench is available under managed-agent project workbench.
+
+#### Examples
+
+```bash
+bl managed-agent playground
+```
+
+```bash
+bl managed-agent playground --agent assistant
+```
+
+```bash
+bl managed-agent playground --file agents.yaml --no-open
+```
+
+### `bl managed-agent project build`
+
+| Field              | Value                                                              |
+| ------------------ | ------------------------------------------------------------------ |
+| **Name**           | `managed-agent project build`                                      |
+| **Description**    | Organize directory source and generate the immutable Publish Build |
+| **Authentication** | No Auth                                                            |
+| **Usage**          | `bl managed-agent project build [--project <directory>]`           |
+
+#### Flags
+
+| Flag                    | Type   | Required | Description                                         |
+| ----------------------- | ------ | -------- | --------------------------------------------------- |
+| `--project <directory>` | string | no       | Directory project root (default: current directory) |
+
+#### Notes
+
+- Build writes local project files without confirmation, including inferred resource associations and migration of plaintext Vault secrets into .env. Use --dry-run to preview without writing. Publish still requires explicit confirmation before remote changes.
+
+#### Examples
+
+```bash
+bl managed-agent project build
+```
+
+```bash
+bl managed-agent project build --dry-run
+```
+
+```bash
+bl managed-agent project build --project ./my-agent
+```
+
+### `bl managed-agent project init`
+
+| Field              | Value                                                       |
+| ------------------ | ----------------------------------------------------------- |
+| **Name**           | `managed-agent project init`                                |
+| **Description**    | Create a directory project or convert the local agents.yaml |
+| **Authentication** | No Auth                                                     |
+| **Usage**          | `bl managed-agent project init [--project <directory>]`     |
+
+#### Flags
+
+| Flag                    | Type   | Required | Description                                                                   |
+| ----------------------- | ------ | -------- | ----------------------------------------------------------------------------- |
+| `--project <directory>` | string | no       | Directory project root (default: ./managed-agent under the current directory) |
+
+#### Notes
+
+- Without --project, creates a managed-agent/ subdirectory. Enter it before running other project commands. Use --project . to initialize in place or convert the current agents.yaml; existing project files are not overwritten.
+- New projects include Skill, File, Vault, and Environment examples under each resource directory's \_examples/. They are not referenced by agent.json and are excluded from Build/Publish. Copy an example outside \_examples/ to enable it, then configure its Agent reference.
+
+#### Examples
+
+```bash
+bl managed-agent project init
+```
+
+```bash
+bl managed-agent project init --project ./my-agent
+```
+
+```bash
+bl managed-agent project init --project .
+```
+
+### `bl managed-agent project publish`
+
+| Field              | Value                                                                                                                |
+| ------------------ | -------------------------------------------------------------------------------------------------------------------- |
+| **Name**           | `managed-agent project publish`                                                                                      |
+| **Description**    | Publish the current directory-project Build and record a version                                                     |
+| **Authentication** | API Key                                                                                                              |
+| **Usage**          | `bl managed-agent project publish [--project <directory>] [--no-refresh] [--concurrency <n>]`                        |
+| **Risk**           | `high`                                                                                                               |
+| **Risk message**   | This publishes the current directory-project Build and may create, update, or delete remote managed Agent resources. |
+
+> **Agent safety:** Never add `--yes` automatically. On `type="requires_confirmation"`, stop and ask for explicit user confirmation of the same action and scope.
+
+#### Flags
+
+| Flag                    | Type   | Required | Description                                         |
+| ----------------------- | ------ | -------- | --------------------------------------------------- |
+| `--project <directory>` | string | no       | Directory project root (default: current directory) |
+| `--no-refresh`          | switch | no       | Skip remote refresh before planning                 |
+| `--concurrency <n>`     | number | no       | Maximum parallel resource operations                |
+| `--yes`                 | switch | no       | Confirm this high-risk operation                    |
+| `--api-key <key>`       | string | no       | API key                                             |
+| `--base-url <url>`      | string | no       | API base URL                                        |
+
+#### Notes
+
+- bl managed-agent supports the Bailian provider only; configurations containing other providers are rejected.
+- Bailian credentials come from bl's auth chain: --api-key > DASHSCOPE_API_KEY > `bl auth login` (active config profile).
+- Resolved credentials are injected into the SDK in-memory and cleared from the environment; they never persist in process env.
+
+#### Examples
+
+```bash
+# Only after explicit user confirmation:
+bl managed-agent project publish --yes
+```
+
+```bash
+# Only after explicit user confirmation:
+bl managed-agent project publish --project ./my-agent --yes
+```
+
+### `bl managed-agent project validate`
+
+| Field              | Value                                                       |
+| ------------------ | ----------------------------------------------------------- |
+| **Name**           | `managed-agent project validate`                            |
+| **Description**    | Validate a directory Agent project                          |
+| **Authentication** | No Auth                                                     |
+| **Usage**          | `bl managed-agent project validate [--project <directory>]` |
+
+#### Flags
+
+| Flag                    | Type   | Required | Description                                         |
+| ----------------------- | ------ | -------- | --------------------------------------------------- |
+| `--project <directory>` | string | no       | Directory project root (default: current directory) |
+
+#### Examples
+
+```bash
+bl managed-agent project validate
+```
+
+```bash
+bl managed-agent project validate --project ./my-agent
+```
+
+### `bl managed-agent project version disable`
+
+| Field              | Value                                                              |
+| ------------------ | ------------------------------------------------------------------ |
+| **Name**           | `managed-agent project version disable`                            |
+| **Description**    | Disable directory project versions                                 |
+| **Authentication** | No Auth                                                            |
+| **Usage**          | `bl managed-agent project version disable [--project <directory>]` |
+
+#### Flags
+
+| Flag                    | Type   | Required | Description                                         |
+| ----------------------- | ------ | -------- | --------------------------------------------------- |
+| `--project <directory>` | string | no       | Directory project root (default: current directory) |
+
+#### Examples
+
+```bash
+bl managed-agent project version disable
+```
+
+```bash
+bl managed-agent project version disable --project ./my-agent
+```
+
+### `bl managed-agent project version enable`
+
+| Field              | Value                                                             |
+| ------------------ | ----------------------------------------------------------------- |
+| **Name**           | `managed-agent project version enable`                            |
+| **Description**    | Enable directory project versions                                 |
+| **Authentication** | No Auth                                                           |
+| **Usage**          | `bl managed-agent project version enable [--project <directory>]` |
+
+#### Flags
+
+| Flag                    | Type   | Required | Description                                         |
+| ----------------------- | ------ | -------- | --------------------------------------------------- |
+| `--project <directory>` | string | no       | Directory project root (default: current directory) |
+
+#### Examples
+
+```bash
+bl managed-agent project version enable
+```
+
+```bash
+bl managed-agent project version enable --project ./my-agent
+```
+
+### `bl managed-agent project version list`
+
+| Field              | Value                                                                                             |
+| ------------------ | ------------------------------------------------------------------------------------------------- |
+| **Name**           | `managed-agent project version list`                                                              |
+| **Description**    | List directory project versions                                                                   |
+| **Authentication** | No Auth                                                                                           |
+| **Usage**          | `bl managed-agent project version list [--project <directory>] [--limit <n>] [--cursor <cursor>]` |
+
+#### Flags
+
+| Flag                    | Type   | Required | Description                                         |
+| ----------------------- | ------ | -------- | --------------------------------------------------- |
+| `--project <directory>` | string | no       | Directory project root (default: current directory) |
+| `--limit <n>`           | number | no       | Maximum versions to return                          |
+| `--cursor <cursor>`     | string | no       | Pagination cursor                                   |
+
+#### Examples
+
+```bash
+bl managed-agent project version list
+```
+
+```bash
+bl managed-agent project version list --limit 20 --output json
+```
+
+### `bl managed-agent project version preview`
+
+| Field              | Value                                                                                          |
+| ------------------ | ---------------------------------------------------------------------------------------------- |
+| **Name**           | `managed-agent project version preview`                                                        |
+| **Description**    | Preview a directory project version                                                            |
+| **Authentication** | No Auth                                                                                        |
+| **Usage**          | `bl managed-agent project version preview --version-id <full-version> [--project <directory>]` |
+
+#### Flags
+
+| Flag                          | Type   | Required | Description                                         |
+| ----------------------------- | ------ | -------- | --------------------------------------------------- |
+| `--project <directory>`       | string | no       | Directory project root (default: current directory) |
+| `--version-id <full-version>` | string | yes      | Full project version ID                             |
+
+#### Examples
+
+```bash
+bl managed-agent project version preview --version-id <full-version>
+```
+
+### `bl managed-agent project version restore`
+
+| Field              | Value                                                                                                        |
+| ------------------ | ------------------------------------------------------------------------------------------------------------ |
+| **Name**           | `managed-agent project version restore`                                                                      |
+| **Description**    | Restore a version to the project working directory                                                           |
+| **Authentication** | No Auth                                                                                                      |
+| **Usage**          | `bl managed-agent project version restore --version-id <full-version> [--project <directory>]`               |
+| **Risk**           | `high`                                                                                                       |
+| **Risk message**   | This restores the full directory source to the working tree. Version history and remote State will not move. |
+
+> **Agent safety:** Never add `--yes` automatically. On `type="requires_confirmation"`, stop and ask for explicit user confirmation of the same action and scope.
+
+#### Flags
+
+| Flag                          | Type   | Required | Description                                         |
+| ----------------------------- | ------ | -------- | --------------------------------------------------- |
+| `--project <directory>`       | string | no       | Directory project root (default: current directory) |
+| `--version-id <full-version>` | string | yes      | Full project version ID                             |
+| `--yes`                       | switch | no       | Confirm this high-risk operation                    |
+
+#### Examples
+
+```bash
+bl managed-agent project version restore --version-id <full-version>
+```
+
+```bash
+# Only after explicit user confirmation:
+bl managed-agent project version restore --version-id <full-version> --yes
+```
+
+### `bl managed-agent project version status`
+
+| Field              | Value                                                             |
+| ------------------ | ----------------------------------------------------------------- |
+| **Name**           | `managed-agent project version status`                            |
+| **Description**    | Show directory project version status                             |
+| **Authentication** | No Auth                                                           |
+| **Usage**          | `bl managed-agent project version status [--project <directory>]` |
+
+#### Flags
+
+| Flag                    | Type   | Required | Description                                         |
+| ----------------------- | ------ | -------- | --------------------------------------------------- |
+| `--project <directory>` | string | no       | Directory project root (default: current directory) |
+
+#### Examples
+
+```bash
+bl managed-agent project version status
+```
+
+```bash
+bl managed-agent project version status --project ./my-agent --output json
+```
+
+### `bl managed-agent project workbench`
+
+| Field              | Value                                                                                 |
+| ------------------ | ------------------------------------------------------------------------------------- |
+| **Name**           | `managed-agent project workbench`                                                     |
+| **Description**    | Launch the directory project Workbench                                                |
+| **Authentication** | API Key                                                                               |
+| **Usage**          | `bl managed-agent project workbench [--project <directory>] [--port <n>] [--no-open]` |
+
+#### Flags
+
+| Flag                    | Type   | Required | Description                                         |
+| ----------------------- | ------ | -------- | --------------------------------------------------- |
+| `--project <directory>` | string | no       | Directory project root (default: current directory) |
+| `--port <n>`            | number | no       | Local port (default: 4848)                          |
+| `--no-open`             | switch | no       | Do not open a browser                               |
+| `--api-key <key>`       | string | no       | API key                                             |
+| `--base-url <url>`      | string | no       | API base URL                                        |
+
+#### Notes
+
+- bl managed-agent supports the Bailian provider only; configurations containing other providers are rejected.
+- Bailian credentials come from bl's auth chain: --api-key > DASHSCOPE_API_KEY > `bl auth login` (active config profile).
+- Resolved credentials are injected into the SDK in-memory and cleared from the environment; they never persist in process env.
+
+#### Examples
+
+```bash
+bl managed-agent project workbench
+```
+
+```bash
+bl managed-agent project workbench --project ./my-agent --no-open
+```
+
 ### `bl managed-agent session archive`
 
 | Field              | Value                                                     |

skills/bailian-protocol/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: bailian-protocol
 metadata:
-  version: "1.20.0"
+  version: "1.23.0"
   requires:
     bins: ["bl"]
 description: >-

skills/bailian-web-search/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: bailian-web-search
 metadata:
-  version: "1.20.0"
+  version: "1.23.0"
   requires:
     bins: ["bl"]
 description: >-