diff --git a/CHANGELOG.md b/CHANGELOG.md
index 1a7bf390..191b7884 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -6,6 +6,34 @@ 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)
 
+## [2.7.3] - 2026-11-03
+
+### Added
+
+- **Custom model import**: Add `bl cmodel base-models/import/get/list/watch/delete` to import Bailian base models as custom models.
+  - `cmodel base-models` lists base models eligible for import (console API; import is currently available in the Beijing region only).
+  - `cmodel import` submits an import job with weight type, display name, and optional bucket/object-key, validating the base model against the console-supported import list before submission.
+  - `cmodel watch --job-id <id>` tracks import job progress with `--follow`, polling interval, and timeout options.
+  - `cmodel get/list/delete` query and remove imported custom models.
+
+### Changed
+
+- **Unified output contract**: Route every command family through one invocation output contract (result / progress / info / warning / debug). Progress and informational messages go to stderr and are hidden in quiet mode; warnings and errors are always preserved. In JSON mode, warnings and errors are emitted as single-line JSON on stderr. Quiet mode no longer trims result fields and gains a `DASHSCOPE_QUIET=1` environment variable.
+- **Advisor thinking disabled**: `bl advisor recommend` LLM calls no longer use thinking mode; intent analysis was the only thinking stage. `--no-thinking` remains accepted as a deprecated no-op for existing scripts.
+
+### Fixed
+
+- **Root help configuration**: Reuse the loaded configuration when rendering root help, preserving parsed flags, Profile selection, and credential precedence.
+- **Duplicate error messages**: Stop repeating the redundant normalization cause in text errors while preserving explicit causes, error codes, JSON details, and verbose stacks.
+- **SDK warnings and empty ASR results**: Preserve SDK warnings and empty ASR task results in managed-agent workflows and `bl speech recognize`.
+- **Finetune fee diagnostics**: `bl finetune fee` now emits gateway diagnostics for fee queries in verbose mode.
+
+### Internal
+
+- Instrument advisor LLM requests for telemetry.
+- Align all command tests and guidance with the output contract; optimize CLI test suites and shard CI test jobs.
+- Clarify quiet output behavior in the CLI and Knowledge Studio READMEs.
+
 ## [2.7.2] - 2026-10-30
 
 ### Added
diff --git a/CHANGELOG.zh.md b/CHANGELOG.zh.md
index bd6abd2e..724d8240 100644
--- a/CHANGELOG.zh.md
+++ b/CHANGELOG.zh.md
@@ -6,6 +6,34 @@
 
 [English](CHANGELOG.md) · [README](README.zh.md) · [参与贡献](CONTRIBUTING.zh.md)
 
+## [2.7.3] - 2026-11-03
+
+### 新增
+
+- **自定义模型导入**：新增 `bl cmodel base-models/import/get/list/watch/delete`，支持将百炼基础模型导入为自定义模型。
+  - `cmodel base-models` 列出可导入的基础模型（控制台接口动态返回；导入当前仅北京地域开放）。
+  - `cmodel import` 提交导入任务，支持权重类型、展示名称与可选 bucket/object-key，提交前按控制台可导入清单校验基础模型。
+  - `cmodel watch --job-id <id>` 跟踪导入任务进度，支持 `--follow`、轮询间隔与超时配置。
+  - `cmodel get/list/delete` 查询与删除已导入的自定义模型。
+
+### 变更
+
+- **统一输出契约**：全部命令统一走本次调用的输出契约（result / progress / info / warning / debug）。进度与过程提示写 stderr（quiet 下隐藏），警告与错误始终保留；JSON 模式下警告/错误以 stderr 单行 JSON 输出。quiet 不再裁剪结果字段，并新增 `DASHSCOPE_QUIET=1` 环境变量。
+- **Advisor 关闭思考模式**：`bl advisor recommend` 的 LLM 调用始终关闭思考模式（意图分析此前为唯一开启思考的阶段）；`--no-thinking` 仍被接受但已无效果，为兼容旧脚本保留。
+
+### 修复
+
+- **根帮助配置复用**：渲染根帮助时复用已加载的配置，保留 flag 解析、Profile 选择与凭证优先级。
+- **重复错误信息**：文本错误不再重复输出归一化 cause，同时保留显式 cause、错误码、JSON 详情与 verbose 堆栈。
+- **SDK 警告与空 ASR 结果**：managed-agent 工作流与 `bl speech recognize` 保留 SDK 警告与空的 ASR 任务结果。
+- **微调费用诊断**：`bl finetune fee` 在 verbose 模式下输出费用查询的 gateway 诊断信息。
+
+### 内部
+
+- Advisor LLM 调用接入请求埋点。
+- 全部命令测试与指引对齐输出契约；优化 CLI 测试套件并对 CI 测试任务分片。
+- 澄清 CLI 与 Knowledge Studio README 中 quiet 行为说明。
+
 ## [2.7.2] - 2026-10-30
 
 ### 新增
diff --git a/docs/agents/auth-change.md b/docs/agents/auth-change.md
index b4462926..bceb6707 100644
--- a/docs/agents/auth-change.md
+++ b/docs/agents/auth-change.md
@@ -29,6 +29,10 @@ defineCommand({ auth }) → runtime/authStage → ctx.client → command.run(ctx
 - `openapi` — 阿里云 OpenAPI 签名域,用 AccessKey ID/Secret 调用 Token Plan 等 OpenAPI
 - `none` — 本地命令、登录/配置类命令、无需 credential 的命令
 
+### 可选 console 副凭证
+
+`auth: "apiKey"` 的命令可用 `optionalConsole: true` 声明 best-effort console 副凭证(例:`cmodel import` 提交前用 console 目录预检基础模型)。runtime 在注入 apiCred 后尽力 `resolveConsole`;解析失败静默跳过,**不得**让 apiKey 主流程失败。此类命令的 help 会额外展示 console 域 flag(`--console-region` / `--console-site` / `--console-switch-agent`),命令捕获 `client.console` 的 AUTH 错误后降级并 warning。纯 console 请求仍必须声明 `auth: "console"`。
+
 ### 多凭证并存
 
 `~/.bailian/config.json` 可同时保存 `api_key`、`access_token` 与 `access_key_*`。登录任一种方式不得删除另一种:
@@ -130,6 +134,7 @@ defineCommand({ auth }) → runtime/authStage → ctx.client → command.run(ctx
   - `globalFlags` 与 `ownFlags` 分流正确
 - [ ] `packages/runtime/src/middleware.ts:authStage`:
   - 根据 `command.auth` 解析 credential 并注入 `ctx.client`
+  - `optionalConsole` 只做 best-effort 副注入;console 解析失败不得影响 apiKey 主流程
   - `settings.dryRun` 下是否允许缺 credential 的策略明确
 - [ ] `packages/runtime/src/error-handler.ts`:
   - AUTH hint 增强使用 `binName`,不要硬编码 `bl`
diff --git a/docs/agents/command-add-remove.md b/docs/agents/command-add-remove.md
index 3cc475a7..c10aa633 100644
--- a/docs/agents/command-add-remove.md
+++ b/docs/agents/command-add-remove.md
@@ -67,6 +67,7 @@ packages/commands/src/index.ts
 - [ ] 新建/删除/移动对应的 `packages/commands/src/commands/<...>.ts`
 - [ ] `defineCommand` 字段使用当前 schema:
   - `auth: "apiKey" | "console" | "openapi" | "none"`
+  - `optionalConsole: true` 仅用于 `auth: "apiKey"` 但需要 best-effort console 预检的命令:runtime 会尽力注入 console 凭证并展示 console 域 flag;console 会话缺失/过期不得阻断主流程,命令必须自行捕获 `client.console` 的 AUTH 错误
   - `flags`(camelCase key,由 runtime 渲染为 kebab-case)
   - `usageArgs`(不含 bin/path 前缀)
   - `exampleArgs`(不含 bin/path 前缀)
diff --git a/docs/research/agent-skill-versioning.md b/docs/research/agent-skill-versioning.md
new file mode 100644
index 00000000..b2395515
--- /dev/null
+++ b/docs/research/agent-skill-versioning.md
@@ -0,0 +1,320 @@
+# Agent 扩展能力单元的版本感知与更新提示
+
+- 调研日期：2026-10-30
+- 调研问题：主流 AI coding agent 怎样给扩展能力单元（skill / plugin / extension / MCP server / rules）声明版本、感知版本和提示更新，以及 agent 能否在会话内知道自己正在用的能力包已经过期。
+- 参照版本：Claude Code CHANGELOG 最新条目为 `2.1.314`；Agent Skills spec 已迁移至 `agentskills.io`；Agent Plugins 标准为 `1.0.0`；MCP 最新协议版本串为 `2026-07-28`。
+
+## TL;DR
+
+各家的做法基本一致：版本声明写在 plugin 或 package 清单里，skill 层没有；更新检查在 CLI 或 app 层做，结果只显示给人，不进模型上下文。Agent Skills 规范只定义 6 个 frontmatter 字段，其中没有 `version`。规范的示例把 `version` 放进自由字典 `metadata`，和本仓库 `tools/sync-skill-metadata.ts` 写 `metadata.version` 的做法相同。这种写法符合规范，但没有任何运行时会读它。
+
+让模型自己感知版本也有可用的路径。Claude Code 官方文档为 SKILL.md 提供了 ``!`<command>`` `` 动态注入语法，可以把实时版本号送进模型上下文；这个语法是 Claude Code 专有的，并且可以用 `disableSkillShellExecution` 关掉。Claude Code 自己也改成通过 tool result 把 artifact 的版本变化告诉模型。反面的情况有两条：官方对 `<system-reminder>` 做了转义防护；伪造「critical update」通知已经是有记录的 skill 投毒手法。
+
+## 横向对比表
+
+| Agent / 生态                    | 扩展单元机制                                           | 有无 `version` 字段                                                                                                                                                 | 运行时是否读取                                                    | 更新提示方式                                                                                                                | 来源                                                                                                                                                                                                                                    |
+| ------------------------------- | ------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| Agent Skills spec（跨厂商规范） | `SKILL.md` frontmatter                                 | 无。只有 `name`/`description`（必填）和 `license`/`compatibility`/`metadata`/`allowed-tools`                                                                        | 否。`metadata` 是自由的字符串键值字典，规范不赋予语义             | 规范不涉及，分发和安装明确不在范围内                                                                                        | [agentskills.io/specification](https://agentskills.io/specification)                                                                                                                                                                    |
+| Claude Code Skills              | `SKILL.md`                                             | 无 `version`；有 `metadata`（"Free-form YAML map"，"Claude Code doesn't act on its contents"）                                                                      | 否（`license`/`compatibility` 也是 "accepted but not acted on"）  | 没有 skill 级更新检查；`/skill-doctor` 只报告未使用的 skill 和 context 成本                                                 | [skills.md](https://docs.claude.com/en/docs/claude-code/skills.md)、[CHANGELOG:3392](https://raw.githubusercontent.com/anthropics/claude-code/main/CHANGELOG.md)                                                                        |
+| Claude Code Plugins             | `plugin.json` 和 `.claude-plugin/marketplace.json`     | 有，均为可选。marketplace 有 entry 级 `version`；`plugin.json` 优先                                                                                                 | 是，也是唯一驱动更新的版本                                        | 后台 auto-update（按 marketplace 开关），以及 `/plugin marketplace update`、`claude plugin update`                          | [marketplace-reference](https://code.claude.com/docs/en/plugins/marketplace-reference)、[host-marketplace](https://code.claude.com/docs/en/plugins/host-marketplace)                                                                    |
+| Agent Plugins 1.0.0（开放标准） | `plugin.json`                                          | 有，非必填（`required` 只有 `$schema` 和 `name`），无格式约束（只是 `"type": "string"`）                                                                            | 标准不规定，分发和安装不在范围内                                  | 标准不涉及                                                                                                                  | [plugin.schema.json](https://agent-plugins.org/schemas/1.0.0/plugin.schema.json)                                                                                                                                                        |
+| OpenAI Codex CLI                | `SKILL.md` frontmatter 和 sidecar `agents/openai.yaml` | 无。frontmatter 只反序列化 `name`/`description`/`metadata`，`metadata` 里只取 `short-description`；sidecar 有 `interface`/`dependencies`/`policy`，也没有 `version` | 否（`metadata` 里除 `short-description` 外的键被 serde 静默丢弃） | 没有扩展单元的更新检查（`config.md` 无相关配置项）                                                                          | [parser.rs](https://github.com/openai/codex/blob/main/codex-rs/skills/src/parser.rs)、[metadata.rs](https://github.com/openai/codex/blob/main/codex-rs/skills/src/metadata.rs)                                                          |
+| Cursor Rules                    | `.cursor/rules/*.mdc`                                  | 无（只有 `description`/`globs`/`alwaysApply`）                                                                                                                      | 否                                                                | 无，靠 git 版本控制和手工修改                                                                                               | [cursor.com/docs/context/rules](https://cursor.com/docs/context/rules)                                                                                                                                                                  |
+| Cursor Plugins                  | `plugin.json` / `.cursor-plugin/plugin.json`           | 有 `version`（示例里有），但清单 "only require a `name`"                                                                                                            | 在清单层读取                                                      | 从 GitHub 导入的可开 "Enable Auto Refresh"（push 触发，最多每 10 分钟一次），也可手动 Refresh；Cloud Agent 运行期间锁定版本 | [cursor.com/docs/plugins](https://cursor.com/docs/plugins)                                                                                                                                                                              |
+| OpenCode                        | Skills 和 plugins（`opencode.json`）                   | 无。skill frontmatter 没有 `version`；plugin 数组只收 npm 包名                                                                                                      | 只用来去重（同名同版本只加载一次）                                | 只针对 CLI 本身：`Updater` 服务有 `disable`/`notify`/`auto` 三种策略（默认 `notify`），另有 `opencode upgrade`              | [docs/skills](https://opencode.ai/docs/skills/)、[updater.ts](https://github.com/sst/opencode/blob/dev/packages/cli/src/services/updater.ts)                                                                                            |
+| Cline                           | Agent Skills                                           | 无。frontmatter 白名单写死为 spec 的 6 个字段                                                                                                                       | 否                                                                | 用 SHA-256 内容哈希代替版本：`skills-lock.json` 的 `computedHash`                                                           | [agent-skill.ts](https://github.com/cline/cline/blob/main/sdk/packages/core/src/extensions/agent-plugin/agent-skill.ts)                                                                                                                 |
+| Continue                        | Hub blocks 和 `config.yaml`                            | 有且必填：`config.yaml` 顶层 `version: z.string()`；block 引用支持 `owner/pkg@version`                                                                              | 是，由 `decodeFullSlug` 解析，未写版本时取 `latest`               | 没有 block 更新通知，靠锁定版本或 `latest` 控制分发                                                                         | [config-yaml/schemas/index.ts](https://github.com/continuedev/continue/blob/main/packages/config-yaml/src/schemas/index.ts)、[slugs.ts](https://github.com/continuedev/continue/blob/main/packages/config-yaml/src/interfaces/slugs.ts) |
+| Aider                           | 没有插件体系                                           | 不适用                                                                                                                                                              | 不适用                                                            | 只针对 CLI 本身：每 24h 查一次 PyPI，经 `io.tool_output` 输出给人看                                                         | [versioncheck.py](https://github.com/Aider-AI/aider/blob/main/aider/versioncheck.py)                                                                                                                                                    |
+| MCP                             | server 实现                                            | 有 `serverInfo.version`，另有独立的 `protocolVersion`                                                                                                               | 是，但只在 client 侧的 `initialize` 握手中使用                    | 协议级版本协商（不匹配时返回 `-32602` `Unsupported protocol version`）                                                      | [lifecycle](https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle)                                                                                                                                                   |
+
+## 各 Agent 机制细节
+
+### 1. Agent Skills 规范：没有 `version` 字段
+
+置信度：文档明确。规范现托管于 `agentskills.io`（Anthropic 仓库的 `spec/agent-skills-spec.md` 已改为一行指向该地址），定义了 6 个 frontmatter 字段：
+
+| 字段            | 必填 | 约束                                                |
+| --------------- | ---- | --------------------------------------------------- |
+| `name`          | 是   | ≤64 字符，小写字母/数字/连字符，须与父目录同名      |
+| `description`   | 是   | ≤1024 字符，须写明做什么和何时用                    |
+| `license`       | 否   | 许可名或随包文件指针                                |
+| `compatibility` | 否   | ≤500 字符，环境需求                                 |
+| `metadata`      | 否   | 自由映射，"a map from string keys to string values" |
+| `allowed-tools` | 否   | 空格分隔，标注 "(Experimental)"                     |
+
+`version` 不在其中。规范只在 `metadata` 的示例里用到 `version`（`version: "1.0"`），并说明这些键 "carries no spec-defined meaning"。`metadata` 的用途是 "Clients can use this to store additional properties not defined by the Agent Skills spec"。
+
+所以我们写的 `metadata.version` 符合规范，也和规范示例一致，但它只是附注数据，没有宿主运行时会读取或比对它。
+
+交叉验证（源码）：`anthropics/skills` 仓库的 20 个 `SKILL.md` 里没有一个含 `version:`（`gh search code 'repo:anthropics/skills filename:SKILL.md "version:"'` 返回 `total_count: 0`）。[template/SKILL.md](https://raw.githubusercontent.com/anthropics/skills/main/template/SKILL.md) 只有 `name` 和 `description`，官方 README 也说 frontmatter "requires only two fields"。
+
+### 2. Claude Code：skill 层没有版本，plugin 层有
+
+置信度：文档明确，并用源码交叉验证。
+
+Skill 侧：[skills.md](https://docs.claude.com/en/docs/claude-code/skills.md) 的 frontmatter 参考表共 21 个字段（`name`/`description`/`when_to_use`/`allowed-tools`/`context`/`agent`/`model`/`hooks`/`paths` 等），没有 `version`。`metadata` 的描述是 "Free-form YAML map for your own key-value data"，文档说 "Claude Code doesn't act on its contents"，非 map 的值会被丢弃。`license` 和 `compatibility` 标为 spec 字段，"accepted but not acted on"。注入模型的 skill 列表也不带版本，description 在列表里按 1,536 字符截断（`/skills` 列表另有 250 字符上限，见 CHANGELOG:7141）。
+
+Plugin 侧的版本有实际作用。[host-marketplace](https://code.claude.com/docs/en/plugins/host-marketplace) 的说明如下：
+
+- 更新的触发条件是 "a user gets a new copy of a plugin only when its computed version changes"。
+- 版本来源顺序是 "That version comes from `plugin.json` first, then from the marketplace entry"。两处都写时静默采用 `plugin.json`，并给出校验告警。
+- 作者可以每次发版都 bump `version`，不 bump 的话 "users stay on their cached copy until the string changes"；也可以不写 `version`，这时 "users track your commits instead"。
+- 例外：本地 marketplace 的 in-place 插件 "isn't controlled by `version`"，每次会话启动都读取当前文件。
+- `command` 源的 copy 模式用内容哈希得出版本："derives the [version] from a hash of the copied files"，"A re-run that produces identical files counts as up to date"。
+
+更新通道（[install](https://code.claude.com/docs/en/plugins/install)）：后台 auto-update 按 marketplace 开关，官方 marketplace 默认开，第三方默认关。`marketplace.json` 里没有字段能打开它，只能由用户在 `/plugin` 里选 "Enable auto-update"，或由管理员在 managed settings 的 `extraKnownMarketplaces` 里设 `"autoUpdate": true`。手动更新用 `/plugin marketplace update <name>` 和 `claude plugin update <plugin>@<marketplace>`，并且 "There's no command that updates every plugin at once"。会话中已加载的版本不会热替换，只会提示 `Plugin updated: <name> · Run /reload-plugins to apply`。
+
+Anthropic 自己的做法（源码）：[anthropics/skills 的 marketplace.json](https://raw.githubusercontent.com/anthropics/skills/main/.claude-plugin/marketplace.json) 把 `version: "1.0.0"` 放在顶层 `metadata` 里，两个 plugin entry（`document-skills`、`example-skills`）都没有 `version`，也就是不写版本、跟随 commit。
+
+附注：marketplace.json 中 `$schema` 指向的 `https://anthropic.com/claude-code/marketplace.schema.json` 曾返回 404（[issue #9686](https://github.com/anthropics/claude-code/issues/9686)，2025-10-16 提交，已关闭）。
+
+### 3. Agent Plugins 1.0.0：有 `version`，非必填，无格式约束
+
+置信度：从 JSON Schema 源码读出。[plugin.schema.json](https://agent-plugins.org/schemas/1.0.0/plugin.schema.json) 的相关部分：
+
+```json
+"version": { "type": "string" },
+...
+"required": ["$schema", "name"],
+"additionalProperties": false
+```
+
+`version` 是普通字符串，没有 semver pattern，也不在 `required` 里（必填的只有 `$schema` 和 `name`）。相比之下 `name` 有 `maxLength: 64` 和严格的 pattern，标准没有给 `version` 规定语义。标准把 "distribution, installation, permissions, user experience" 留给各客户端。Cursor 和 Claude Code 都采用这份 plugin 清单标准。
+
+### 4. OpenAI Codex CLI：没有 `version`，有 `dependencies` 声明
+
+置信度：源码读出。`developers.openai.com/codex/skills` 返回 403，仓库内的 `docs/skills.md` 只有一行外链。Codex 的解析逻辑在开源的 Rust 代码里，以下结论来自结构体定义。
+
+[`codex-rs/skills/src/parser.rs`](https://github.com/openai/codex/blob/main/codex-rs/skills/src/parser.rs) 的 frontmatter 结构体：
+
+```rust
+struct SkillFrontmatter {
+    name: Option<String>,
+    description: Option<String>,
+    metadata: SkillFrontmatterMetadata,
+}
+struct SkillFrontmatterMetadata {
+    #[serde(default, rename = "short-description")]
+    short_description: Option<String>,
+}
+```
+
+校验后的 `ParsedSkillFrontmatter` 只有 `name`、`description`、`short_description` 三个成员。
+
+Codex 的 SKILL.md frontmatter 没有 `version`。它会解析 `metadata`，但只取 `short-description`。结构体没有 `deny_unknown_fields`，所以我们写的 `metadata.version` 和 `metadata.requires` 会被 serde 静默丢弃，不报错也不警告。
+
+Codex 还有一个 sidecar 元数据文件。[`codex-rs/skills/src/metadata.rs`](https://github.com/openai/codex/blob/main/codex-rs/skills/src/metadata.rs) 的 doc comment 说明它解析 `agents/openai.yaml`，顶层有三个键：
+
+- `interface`：skill 的接口资产
+- `dependencies`：其中 `dependencies.tools[]` 可以声明所需的工具或 MCP，字段为 `type`/`value`/`description`/`transport`/`command`/`url`/`oauth`
+- `policy`：`allow_implicit_invocation` 和 `products`
+
+`dependencies` 里也没有版本或版本区间字段，只有 `MAX_DEPENDENCY_*_LEN` 这类长度限制。
+
+`dependencies.tools` 让 skill 用结构化字段声明自己需要的外部工具，用意和我们的 `metadata.requires.bins: ["bl"]` 相同。它只声明需要哪个工具，没有版本要求。要表达「需要 ≥2.7.1 的 `bl`」，已调研的宿主都没有原生字段，最接近的是 VS Code 的 `engines`（见下节）。
+
+### 5. Cursor：rules 没有版本，靠 git 管理；plugins 有版本，靠 push 刷新
+
+置信度：文档明确。[rules 文档](https://cursor.com/docs/context/rules) 的 `.mdc` frontmatter 只有 `description`、`globs`、`alwaysApply` 三个字段，全文没有 version、schema-version 或 revision 字段，也没有自动更新。文档给的办法是：project rules "version-controlled"（交给 git），发现 agent 出错后手工修改；"Rules aren't imported on their own"，跨仓库分发要打包成 plugin。
+
+[plugins 文档](https://cursor.com/docs/plugins) 里的两种清单格式（Agent Plugins 开放标准的 `plugin.json` 和 Cursor 原生的 `.cursor-plugin/plugin.json`）示例都含 `"version": "1.0.0"`，但 Cursor 格式 "only require a `name`"。更新由仓库事件触发，不比对版本串：从 GitHub 导入的插件可以开 "Enable Auto Refresh"（需要 Cursor GitHub App，推送后更新，最多每 10 分钟一次），也可以手动点 Refresh。运行中的 Cloud Agent "keeps the plugin versions it started with"，要用新版本得新起一个 agent。
+
+### 6. OpenCode / Cline / Continue / Aider
+
+置信度：源码读出（部分文档站返回 403）。
+
+- OpenCode：skill frontmatter 与 spec 的六个字段一致，没有 `version`，且 "Unknown frontmatter fields are ignored"。plugin 数组只接受 npm 包名（`["opencode-wakatime", "@my-org/custom-plugin"]`），不支持 `@version` 锁定，版本只用来去重（"Duplicate npm packages with the same name and version are loaded once"）。更新机制只管 CLI 本身：[updater.ts](https://github.com/sst/opencode/blob/dev/packages/cli/src/services/updater.ts) 实现了 `disable`/`notify`/`auto` 三种策略，默认 `notify`，日志记录 `{current, latest, action}`。
+- Cline：[agent-skill.ts](https://github.com/cline/cline/blob/main/sdk/packages/core/src/extensions/agent-plugin/agent-skill.ts) 把 frontmatter 白名单写死为 spec 的 6 个字段，`AgentSkillMetadata` 没有 `version` 成员。Cline 用内容哈希代替版本：`skills-lock.json` 记录 `computedHash`(SHA-256) 以及 `source`/`sourceType`/`skillPath`，其中的 `version: 1` 是 lockfile 的格式版本。这套机制检测漂移和完整性，不负责通知更新。
+- Continue：四者中只有它有扩展单元的版本体系。`config.yaml` 顶层的 `version: z.string()` 必填；block 引用经 `FullSlug = {ownerSlug, packageSlug, versionSlug}` 解析，`encodeFullSlug` 生成 `owner/pkg@version`，`decodeFullSlug` 在未写版本时取 `VirtualTags.Latest`。版本由配置加载器（CLI 或扩展层）使用，不注入模型 prompt。
+- Aider：没有插件体系，[versioncheck.py](https://github.com/Aider-AI/aider/blob/main/aider/versioncheck.py) 是一个完整的 CLI 层实现。它用 `~/.aider/caches/versioncheck` 的 mtime 记录状态（不存内容），每 24h 检查一次，延迟导入 `requests`（"To keep startup fast"），查 PyPI JSON 后用 `packaging.version.parse` 比较；`finally` 块无论成败都更新时间戳，避免失败后反复重试。输出走 `io.tool_output` / `io.tool_warning`，给人看，不进模型上下文。
+
+## 其他生态的类似机制
+
+### npm update-notifier
+
+置信度：文档明确。[update-notifier readme](https://raw.githubusercontent.com/yeoman/update-notifier/main/readme.md) 的设计大多可以直接用在我们这里：
+
+- 异步加缓存，不拖慢启动：超过间隔才在后台查 npm 并写盘，下次运行从缓存读出结果填进 `.update`，"so startup speed isn't affected"。检查跑在 unref'ed 子进程里，主进程 `process.exit` 后也能完成。
+- 状态文件在 `~/.config/configstore/update-notifier-[module].json`，路径由 `notifier.config.path` 暴露。
+- `updateCheckInterval` 默认 1 天；`distTag` 默认 `latest`。
+- `notify()` 只在有更新且进程是 TTY 时打印，默认推迟到进程退出时显示（boxen 黄框）。
+- 关闭方式有三种：配置里的 `optOut: true`、`NO_UPDATE_NOTIFIER` 环境变量、`--no-update-notifier` flag。CI 和 `NODE_ENV=test` 下直接跳过。
+- 第一次运行不提示（间隔还没到），写 e2e 断言时要考虑这一点。
+
+### VS Code extension：`version` 和 `engines`
+
+置信度：文档明确。[extension-manifest](https://code.visualstudio.com/api/references/extension-manifest) 里 `version` 必填（"SemVer compatible version"）；`engines` 也必填，至少包含 `vscode` 键，"Cannot be `*`"，例如 `^0.10.5` 表示最低兼容版本。前者是扩展自己的版本，后者声明它兼容的宿主版本。我们的 skill 需要声明「需要多新的 `bl`」，和 `engines` 是同一类需求。Agent Skills 的 `compatibility` 是自由文本，而且 "not acted on"，起不到同样的作用。
+
+### MCP：版本协商在 client 层完成，不经过模型
+
+置信度：规范原文和 schema 源码。[lifecycle 规范](https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle) 区分两种版本：
+
+- `protocolVersion`（如 `"2025-06-18"`）：client 发送自己支持的最新版本；server 支持则 MUST 回同一版本，否则 MUST 回一个自己支持的版本；client 无法处理时 SHOULD 断开。HTTP 下后续请求要带 `MCP-Protocol-Version` 头。版本不匹配的错误示例是 `-32602` 加 `"Unsupported protocol version"`，`data` 里列出 `supported` 和 `requested`。
+- `serverInfo.version`（如 `"1.0.0"`）：实现本身的版本，与 `protocolVersion` 无关。`clientInfo` 和 `serverInfo` 结构相同，都是 `name`/`title`/`version`。
+
+tools 规范的消息流程图（[tools.mdx](https://raw.githubusercontent.com/modelcontextprotocol/modelcontextprotocol/blob/main/docs/specification/2025-06-18/server/tools.mdx)，源码读出）列出了 `LLM` 参与者，但 Updates 阶段只有 `Server--)Client` 和随后的 `Client->>Server: tools/list`，LLM 不参与版本和变更的传递。规范里唯一面向模型的字段是 `InitializeResult.instructions`，[schema.ts](https://raw.githubusercontent.com/modelcontextprotocol/modelcontextprotocol/blob/main/schema/2025-06-18/schema.ts) 的描述是 "like a 'hint' to the model. For example, this information MAY be added to the system prompt"，是否加入由 client 决定。server 可以把版本号写进 `instructions`，规范既没有建议也没有禁止。
+
+### Homebrew 与 gh CLI
+
+置信度：源码读出。[cli/cli 的 cmd.go](https://raw.githubusercontent.com/cli/cli/trunk/internal/ghcmd/cmd.go) 用 `fmt.Fprintf(stderr, ...)` 加 ANSI 颜色把更新提示写到 stderr，给人看，不进入模型上下文。[update.go](https://raw.githubusercontent.com/cli/cli/trunk/internal/update/update.go) 只在有人坐在终端前时提示，条件是 `!ci.IsCI() && IsTerminal(os.Stdout) && IsTerminal(os.Stderr)`，另外可以用 `GH_NO_EXTENSION_UPDATE_NOTIFIER`、`CODESPACES` 关闭。Homebrew 用 `HOMEBREW_NO_AUTO_UPDATE` 控制自动更新，用 `brew outdated` 单独查询。
+
+## 会话内版本感知的先例
+
+### (a) 把版本注入模型上下文：Claude Code 有官方语法
+
+置信度：文档明确。Claude Code [skills.md 第 705 行 "Inject dynamic context"](https://docs.claude.com/en/docs/claude-code/skills.md) 原文：
+
+> The `` !`<command>` `` syntax runs shell commands before the skill content is sent to Claude. The command output replaces the placeholder, so Claude receives actual data, not the command itself.
+
+官方示例在 SKILL.md 正文里写 ``- PR diff: !`gh pr diff` ``。内联写法有位置要求："The inline form is only recognized when `!` appears at the start of a line or immediately after whitespace"；多行命令用 ` ```! ` 开头的围栏代码块。
+
+文档没有给这个语法本身设版本下限。`v2.1.228+` 只针对另一条安全限制：从 claude.ai 账号同步来的 skill 不在本机执行这些命令，原文 "This restriction requires Claude Code v2.1.228 or later"。
+
+settings 里的 `disableSkillShellExecution` 可以关掉这个功能，开启后每条命令被替换成 policy-disabled 占位符（bundled 和 managed skill 不受影响）。依赖这个语法的方案要处理占位符原样留在上下文里的情况。
+
+在 SKILL.md 正文写 ``!`bl --version`` `` 可以把真实版本号送进模型上下文，可以放在 `bailian-protocol` 现有的「版本预检」环节。
+
+### (b) Claude Code 通过 tool result 把版本变化告诉模型
+
+置信度：官方 CHANGELOG 原文。CHANGELOG 第 2850 行：
+
+> Changed artifact watching in local sessions: a new version published elsewhere no longer starts a turn; Claude learns of it from a later Artifact tool result
+
+这次改动让版本变化的信号随 tool result 进入模型上下文，和我们想做的事属于同一类，只是对象是 artifact。信号附在后续的 tool result 里，不再单独发起一个 turn 打断用户。
+
+### (c) skill 写入自身版本号、让模型自查是否过期：没有找到公开先例
+
+置信度：基于代码检索，查询语句如下。`gh search code '"your version is outdated" filename:SKILL.md'` 只有 4 条结果，其中两条是提示注入攻击样本（`antgroup/Agent3Sigma-Canary/assets/skill_poison/trusted-agent-deception/critical-update/SKILL.md`），说明伪造「critical update」通知已被收录为一种 skill 投毒手法。
+
+唯一的正面样本是 [ykdojo/claude-code-tips 的 skills/version-check/SKILL.md](https://raw.githubusercontent.com/ykdojo/claude-code-tips/main/skills/version-check/SKILL.md)（源码读出）。它让 agent 执行 `claude --version` 和 `npm view @anthropic-ai/claude-code dist-tags --json`，对照 changelog 后给出 update、pin 或 stay 的建议。它检查的是宿主 CLI 的版本，由用户主动触发，不检查 skill 自身的版本。
+
+更宽的查询（`"check for updates"`、`"--version" "outdated"`）有数千条结果，基本是讲依赖管理的 skill，和 skill 自身的版本感知无关。
+
+公开生态里没有可以照搬的做法，安全上的风险见 D2。
+
+## 对我们的启示
+
+先确认几条事实：
+
+1. 我们的 `metadata.version` 符合规范，但没有宿主读取 `version`，各宿主忽略的方式不同。Claude Code 文档说 `metadata` "doesn't act on its contents"；Codex 的 `SkillFrontmatterMetadata` 会解析 `metadata`，但只取 `short-description`，`version` 和 `requires` 被 serde 静默丢弃，不报错也不警告；Cline 和 OpenCode 把 frontmatter 白名单写死。目前这个字段只有人和我们自己的工具会看。
+2. 宿主装载 skill 都是复制文件或 clone，版本感知放在 plugin/marketplace 层。我们用 `bl skill init` 自己做安装，所以能在安装器里做宿主做不到的检查。
+3. 本仓库已有两处可以复用：`bailian-protocol` 负责的「版本预检」，以及 `packages/runtime/src/middleware.ts` 里的 update middleware。
+4. 按 `skill-change.md`，纳入 Git 的 `reference/` 不写包版本号（channel 发布会临时 bump 成带时间戳的 beta 版本，否则每次发布都会产生生成物 diff）。把版本写进生成物的方案都要避开这条。
+
+下面 4 个方向可以组合使用。
+
+### D1（推荐，风险低）：`bl` 在 CLI 层检查版本，提示走 agent 读得到的通道
+
+做法：在 runtime 的 update middleware 旁加一个 skill 版本比对，比较已安装 skill 的 `metadata.version` 和当前 `bl` 的版本（两者都来自 `packages/cli/package.json`），不一致时输出一行提示，建议执行 `bl skill update`。
+
+先例：这是最常见的做法。update-notifier（异步、缓存、TTY 判断、多种关闭方式），Aider `versioncheck.py`（mtime 记录状态、24h 间隔、`finally` 防止重试风暴），gh CLI（写 stderr、判断 CI 和 TTY），OpenCode `Updater`（三种策略，默认 `notify`）。
+
+优点：不往 prompt 里加内容，没有投毒入口，有成熟实现可参考，结构和现有 update middleware 相同。
+
+和先例的差别：上面的先例都只在 TTY 下给人看。我们的用户主要是 agent，agent 通常不在 TTY 下运行，照搬 TTY 判断的话，agent 场景下永远看不到提示。建议 TTY 下输出给人看的横幅，非 TTY 下输出一行格式固定、机器可读的前缀（比如写到 stderr 的单行 `notice:`），agent 能读到，stdout 的结构化输出也不受影响。
+
+风险：stderr 多出一行可能影响现有 e2e 断言（可参考 update-notifier 首次运行不提示的处理）；需要提供 env var 关闭开关，并在 CI 下跳过。
+
+### D2：在 SKILL.md 正文用 ``!`bl --version`` `` 做运行时版本预检
+
+做法：在 `bailian-protocol` 的版本预检段落写 ``!`bl --version`` ``（或专用的 `bl skill doctor --format json`），让模型在执行前拿到实际的宿主版本，和 skill 正文或 `metadata` 里的期望版本比对。
+
+先例：Claude Code 官方文档的 "Inject dynamic context"；Claude Code 通过 tool result 把 artifact 版本变化告诉模型（CHANGELOG:2850）。
+
+优点：只有这个方向能让模型在会话中、执行命令前知道版本；版本信息来自我们自己的命令输出；和 `bailian-protocol` 现有的职责一致。
+
+风险：
+
+- 可移植性差。这是 Claude Code 专有语法。OpenCode 说明了 "Unknown frontmatter fields are ignored"，但不保证正文语法；Cursor 和 Codex 没有这个语法，``!`...`` `` 会作为普通文本进入上下文，模型可能把它当成要执行的指令。需要兜底的措辞。
+- 可以被关掉。即使在 Claude Code 里，开启 `disableSkillShellExecution` 后命令也会变成 policy-disabled 占位符。版本预检不能是硬依赖，看到占位符时应跳过版本检查，不报错，也不猜测。
+- 安全。伪造「critical update」是有记录的 skill 投毒手法。版本结论只能来自我们自己的命令输出，不能来自正文里的文字断言。提示语气要平和，免得用户对更新警告习以为常。
+- 从 claude.ai 同步的 skill 不执行该语法（v2.1.228+ 起生效）。
+
+### D3：提供 `.claude-plugin/marketplace.json` 分发通道，接入 plugin 层
+
+做法：在 `bl skill init` 之外，再以 Claude Code plugin marketplace（或 Agent Plugins 1.0.0 `plugin.json`）的形式分发 `bailian-*` skills，使用宿主自带的版本和更新机制。
+
+先例：这是 Claude Code 唯一根据版本触发更新的路径（"a user gets a new copy only when its computed version changes"）；Cursor 也支持 Agent Plugins `plugin.json`；Anthropic 的 `anthropics/skills` 就是这样分发的。
+
+优点：直接用宿主的后台 auto-update 和 `claude plugin update`；这一层的 `version` 会被运行时读取；同一份 `plugin.json` 也能给 Cursor 用。
+
+风险与注意事项：
+
+- 第三方 marketplace 的 auto-update 默认关闭，`marketplace.json` 里也没有字段能打开，只能用户手动开，或管理员通过 managed settings 下发。发布之后不会自动到达用户机器。
+- 官方自己不写 entry `version`，跟随 commit。如果我们写了 `version`，每次发版都必须 bump，否则用户会一直停在缓存副本上。这和我们七包锁步版本的习惯一致，但要加进发布清单。
+- 只覆盖 Claude Code 和 Cursor 这类宿主，覆盖不了全部安装场景。
+- 和现有的 `bl skill init/update` 并存会形成两条安装通道，需要定好优先级，否则会出现以哪边版本为准的冲突。
+
+### D4：用内容哈希检测漂移，发现用户改过的本地 skill
+
+做法：参考 Cline 的 `skills-lock.json`，在 `bl skill` 安装时写一份 lock（`computedHash`、source、path），`bl skill update` 和 `doctor` 用哈希判断本地文件是否与发布版一致、用户是否改过。
+
+先例：Cline `skills-lock.json` 的 `computedHash`(SHA-256)（源码读出）；Claude Code 对 `command` 源插件 "derives the version from a hash of the copied files"，且 "identical files counts as up to date"。
+
+优点：比版本串可靠，能发现用户本地改动、安装中途失败、漏装 `bailian-protocol` 这些我们遇到过的问题；可以和 `metadata.version` 配合使用；不依赖宿主功能。
+
+风险：需要额外的状态文件，也要保证生成结果稳定。`reference/` 生成物在 channel 发布时会临时带上 beta 版本号，哈希计算必须排除含版本号的内容，否则每次发布所有文件都会被判为漂移。
+
+### 建议
+
+D1 和 D4 作为基础，它们不依赖宿主，也不往 prompt 里加内容。D2 作为 `bailian-protocol` 的增强，它是唯一能在会话内感知版本的方向。D3 等看过 Claude Code 的安装占比再决定。D1 要针对「用户是 agent、常在非 TTY 下运行」这一点设计，不能沿用业界只在 TTY 下提示的默认做法。
+
+## 来源清单
+
+### 一手来源（官方文档 / 官方仓库源码 / 协议规范 / 官方 CHANGELOG）
+
+#### 规范
+
+- [Agent Skills 规范](https://agentskills.io/specification)：6 个 frontmatter 字段，无 `version`
+- [anthropics/skills spec/agent-skills-spec.md](https://raw.githubusercontent.com/anthropics/skills/main/spec/agent-skills-spec.md)：只指向上述地址
+- [anthropics/skills template/SKILL.md](https://raw.githubusercontent.com/anthropics/skills/main/template/SKILL.md)
+- [anthropics/skills .claude-plugin/marketplace.json](https://raw.githubusercontent.com/anthropics/skills/main/.claude-plugin/marketplace.json)：顶层 `metadata.version`，entry 无 version
+- [Agent Plugins 1.0.0 plugin.schema.json](https://agent-plugins.org/schemas/1.0.0/plugin.schema.json)：`required: ["$schema","name"]`
+- [agent-plugins.org 总览](https://agent-plugins.org/)
+
+#### Claude Code
+
+- [Skills 文档（frontmatter 表和 "Inject dynamic context" L705）](https://docs.claude.com/en/docs/claude-code/skills.md)
+- [Marketplace reference](https://code.claude.com/docs/en/plugins/marketplace-reference)
+- [Create a marketplace](https://code.claude.com/docs/en/plugins/create-marketplace)
+- [Host a marketplace: Keep users up to date](https://code.claude.com/docs/en/plugins/host-marketplace)
+- [Install and manage plugins](https://code.claude.com/docs/en/plugins/install)
+- [CHANGELOG（v2.1.314；L2850 artifact 版本变化、L1312 system-reminder 转义、L3392 /skill-doctor、L4207 后台通知）](https://raw.githubusercontent.com/anthropics/claude-code/main/CHANGELOG.md)
+- [issue #9686：marketplace schema 404](https://github.com/anthropics/claude-code/issues/9686)
+
+#### MCP
+
+- [Lifecycle 规范（protocolVersion 协商和 serverInfo.version）](https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle)
+- [schema.ts：`instructions` 的 "MAY be added to the system prompt"](https://raw.githubusercontent.com/modelcontextprotocol/modelcontextprotocol/blob/main/schema/2025-06-18/schema.ts)
+- [tools.mdx：Updates 流程图不经过 LLM](https://raw.githubusercontent.com/modelcontextprotocol/modelcontextprotocol/blob/main/docs/specification/2025-06-18/server/tools.mdx)
+
+#### 其他 agent
+
+- [Cursor Rules](https://cursor.com/docs/context/rules)
+- [Cursor Plugins](https://cursor.com/docs/plugins)
+- [OpenAI Codex codex-rs/skills/src/parser.rs（frontmatter 结构体，无 version）](https://github.com/openai/codex/blob/main/codex-rs/skills/src/parser.rs)
+- [OpenAI Codex codex-rs/skills/src/metadata.rs（sidecar `agents/openai.yaml`：interface/dependencies/policy）](https://github.com/openai/codex/blob/main/codex-rs/skills/src/metadata.rs)
+- [OpenAI Codex docs/config.md](https://raw.githubusercontent.com/openai/codex/main/docs/config.md)
+- [OpenAI Codex docs/skills.md（只有一行外链，正文在返回 403 的开发者站）](https://raw.githubusercontent.com/openai/codex/main/docs/skills.md)
+- [OpenCode Skills](https://opencode.ai/docs/skills/) / [Plugins](https://opencode.ai/docs/plugins/) / [updater.ts](https://github.com/sst/opencode/blob/dev/packages/cli/src/services/updater.ts) / [upgrade.ts](https://github.com/sst/opencode/blob/dev/packages/cli/src/commands/handlers/upgrade.ts) / [version.ts](https://github.com/sst/opencode/blob/dev/packages/cli/src/version.ts)
+- [Cline agent-skill.ts（frontmatter 白名单）](https://github.com/cline/cline/blob/main/sdk/packages/core/src/extensions/agent-plugin/agent-skill.ts) / [skills-lock.json](https://github.com/cline/cline/blob/main/apps/vscode/skills-lock.json)
+- [Continue config-yaml schemas/index.ts](https://github.com/continuedev/continue/blob/main/packages/config-yaml/src/schemas/index.ts) / [slugs.ts](https://github.com/continuedev/continue/blob/main/packages/config-yaml/src/interfaces/slugs.ts) / [reference 文档](https://docs.continue.dev/reference)
+- [Aider versioncheck.py](https://github.com/Aider-AI/aider/blob/main/aider/versioncheck.py)
+
+#### 类似机制
+
+- [update-notifier readme](https://raw.githubusercontent.com/yeoman/update-notifier/main/readme.md)
+- [VS Code extension manifest（version 和 engines.vscode）](https://code.visualstudio.com/api/references/extension-manifest)
+- [gh CLI internal/ghcmd/cmd.go](https://raw.githubusercontent.com/cli/cli/trunk/internal/ghcmd/cmd.go) / [internal/update/update.go](https://raw.githubusercontent.com/cli/cli/trunk/internal/update/update.go)
+
+### 二手来源（只作模式举证，非权威）
+
+- [ykdojo/claude-code-tips skills/version-check/SKILL.md](https://raw.githubusercontent.com/ykdojo/claude-code-tips/main/skills/version-check/SKILL.md)：社区个人仓库，是「skill 指导 agent 自查宿主版本」的唯一正面样本
+- [antgroup/Agent3Sigma-Canary skill_poison/.../critical-update/SKILL.md](https://github.com/antgroup/Agent3Sigma-Canary)：第三方安全测试集，用来说明伪造更新通知是已知的投毒手法
+
+### 未能获取或未找到的证据
+
+- `developers.openai.com/codex/skills` 返回 403。Codex 的结论来自 `codex-rs/skills/src/{parser,metadata}.rs` 源码，不依赖该文档站（见第 4 节）。
+- `docs.claude.com/en/docs/claude-code/{skills,plugin-marketplaces,cli-reference}` 的 HTML 页返回 403，改用内容相同的 `.md` 原文和 `code.claude.com`；`skills.md` 已于 2026-10-30 复核。
+- `agentskills.io/schemas/**` 返回 404，没有机器可读的 skill JSON Schema。「规范有 6 个字段」只有文档文字支撑，没有 schema；Claude Code 文档、Cline 和 OpenCode 的字段白名单源码、`anthropics/skills` 检索 0 条结果，三者结论一致。
+- Goose：`block.github.io/goose/docs/guides/recipes/` 返回 404，没有取得一手证据，所以 Goose 没有列入对比表。
diff --git a/packages/bailian-kb-dsh/package.json b/packages/bailian-kb-dsh/package.json
index 234b1654..6d3a08a8 100644
--- a/packages/bailian-kb-dsh/package.json
+++ b/packages/bailian-kb-dsh/package.json
@@ -1,6 +1,6 @@
 {
   "name": "bailian-kb-dsh",
-  "version": "2.7.2",
+  "version": "2.7.3",
   "description": "Bailian knowledge-base tools for DeepSeek Harness: kb_search and kb_chat over the DashScope RAG API, plus the bl CLI management skill.",
   "keywords": [
     "alibaba-cloud",
diff --git a/packages/bailian-memo-dsh/package.json b/packages/bailian-memo-dsh/package.json
index b7c30f21..aae9017d 100644
--- a/packages/bailian-memo-dsh/package.json
+++ b/packages/bailian-memo-dsh/package.json
@@ -1,6 +1,6 @@
 {
   "name": "bailian-memo-dsh",
-  "version": "2.7.2",
+  "version": "2.7.3",
   "description": "Bailian personal memory for DeepSeek Harness: automatic recall, silent curation, and user-controlled remember/forget.",
   "keywords": [
     "alibaba-cloud",
diff --git a/packages/cli/package.json b/packages/cli/package.json
index 8646e251..eba18a8a 100644
--- a/packages/cli/package.json
+++ b/packages/cli/package.json
@@ -1,6 +1,6 @@
 {
   "name": "bailian-cli",
-  "version": "2.7.2",
+  "version": "2.7.3",
   "description": "CLI for Aliyun Model Studio (DashScope) AI Platform.",
   "keywords": [
     "agent",
diff --git a/packages/cli/src/commands.ts b/packages/cli/src/commands.ts
index 39183430..79d07710 100644
--- a/packages/cli/src/commands.ts
+++ b/packages/cli/src/commands.ts
@@ -112,6 +112,12 @@ import {
   modelList,
   modelSearch,
   modelCode,
+  cmodelImport,
+  cmodelGet,
+  cmodelList,
+  cmodelWatch,
+  cmodelDelete,
+  cmodelBaseModels,
   workspaceList,
   quotaList,
   quotaUpdate,
@@ -446,6 +452,12 @@ export const commands: Record<string, AnyCommand> = {
   "model list": modelList,
   "model search": modelSearch,
   "model code": modelCode,
+  "cmodel import": cmodelImport,
+  "cmodel get": cmodelGet,
+  "cmodel list": cmodelList,
+  "cmodel watch": cmodelWatch,
+  "cmodel delete": cmodelDelete,
+  "cmodel base-models": cmodelBaseModels,
   "workspace list": workspaceList,
   "quota list": quotaList,
   "quota update": quotaUpdate,
diff --git a/packages/commands/package.json b/packages/commands/package.json
index 335f8a3f..06137c08 100644
--- a/packages/commands/package.json
+++ b/packages/commands/package.json
@@ -1,6 +1,6 @@
 {
   "name": "bailian-cli-commands",
-  "version": "2.7.2",
+  "version": "2.7.3",
   "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": {
diff --git a/packages/commands/src/commands/cmodel/base-models.ts b/packages/commands/src/commands/cmodel/base-models.ts
new file mode 100644
index 00000000..109ea28d
--- /dev/null
+++ b/packages/commands/src/commands/cmodel/base-models.ts
@@ -0,0 +1,53 @@
+import {
+  defineCommand,
+  effectiveConsoleGatewayConfig,
+  listSupportImportModels,
+  type FlagsDef,
+} from "bailian-cli-core";
+
+export default defineCommand({
+  defaultOutput: "json",
+  description: {
+    "en-US": "List base models that can be imported as custom models",
+    "zh-CN": "列出可导入为自定义模型的基础模型",
+  },
+  auth: "console",
+  usageArgs: "",
+  flags: {} satisfies FlagsDef,
+  exampleArgs: ["", "--console-region cn-beijing", "--output json"],
+  notes: [
+    {
+      "en-US":
+        "The list is returned dynamically by the console API and may change between releases. Import is currently available only in the Beijing region.",
+      "zh-CN": "清单由控制台接口动态返回，可能随版本更新调整。导入当前仅北京地域开放。",
+    },
+  ],
+  async run(ctx) {
+    const { settings } = ctx;
+
+    if (settings.dryRun) {
+      const resultData = {
+        action: "cmodel.baseModels",
+        ...effectiveConsoleGatewayConfig(settings),
+      };
+      ctx.output.result(
+        resultData,
+        ctx.settings.output === "text" ? { text: JSON.stringify(resultData, null, 2) } : undefined,
+      );
+      return;
+    }
+
+    const supported = await listSupportImportModels(ctx.client);
+    const items = supported.list.map((entry) => ({
+      model: entry.model ?? "",
+      name: entry.name ?? "",
+      support_train_types: entry.supportTrainTypes ?? [],
+      post_process_capabilities: entry.postProcessCapabilities ?? null,
+    }));
+    const resultData = { items, total: supported.total ?? items.length };
+    ctx.output.result(
+      resultData,
+      ctx.settings.output === "text" ? { text: JSON.stringify(resultData, null, 2) } : undefined,
+    );
+  },
+});
diff --git a/packages/commands/src/commands/cmodel/delete.ts b/packages/commands/src/commands/cmodel/delete.ts
new file mode 100644
index 00000000..d43da872
--- /dev/null
+++ b/packages/commands/src/commands/cmodel/delete.ts
@@ -0,0 +1,60 @@
+import { defineCommand, deleteCustomModelImport } from "bailian-cli-core";
+
+import { IMPORT_JOB_ID_FLAGS, PUBLIC_DASHSCOPE_NOTE, buildImportJobOutput } from "./shared.ts";
+
+export default defineCommand({
+  defaultOutput: "json",
+  description: {
+    "en-US": "Delete a custom model and its Bailian-side import record",
+    "zh-CN": "删除自定义模型及其百炼侧导入记录",
+  },
+  auth: "apiKey",
+  risk: {
+    level: "high",
+    message: {
+      "en-US":
+        "This permanently deletes the Bailian-side import record and imported model copy; it cannot be undone. OSS source files are not modified.",
+      "zh-CN": "该操作会永久删除百炼侧导入记录及平台侧导入副本，且无法撤销；不会修改 OSS 源文件。",
+    },
+  },
+  usageArgs: "--job-id <id>",
+  flags: IMPORT_JOB_ID_FLAGS,
+  exampleArgs: ["--job-id 937b-xxxx", "--job-id 937b-xxxx --dry-run", "--job-id 937b-xxxx --yes"],
+  notes: [
+    PUBLIC_DASHSCOPE_NOTE,
+    {
+      "en-US":
+        "Irreversible: the import record and imported model are removed from Bailian and must be imported again to restore. OSS source files are untouched.",
+      "zh-CN": "不可撤销：导入记录与导入模型会从百炼删除，需重新导入才能恢复；OSS 源文件不受影响。",
+    },
+    {
+      "en-US":
+        "Only SUCCESSED or FAILED jobs can be deleted; deleting a RUNNING job is rejected by the service (OperationDenied) and the error is surfaced verbatim.",
+      "zh-CN":
+        "仅 SUCCESSED 或 FAILED 状态的任务可删除；删除 RUNNING 任务会被服务端拒绝（OperationDenied），错误原样透出。",
+    },
+  ],
+  async run(ctx) {
+    const { settings, flags } = ctx;
+    const jobId = flags.jobId;
+
+    if (settings.dryRun) {
+      const resultData = { action: "cmodel.delete", job_id: jobId };
+      ctx.output.result(
+        resultData,
+        ctx.settings.output === "text" ? { text: JSON.stringify(resultData, null, 2) } : undefined,
+      );
+      return;
+    }
+
+    const response = await deleteCustomModelImport(ctx.client, jobId);
+    const job = response.output ?? response.data;
+    const resultData = job
+      ? { ...buildImportJobOutput(job, jobId), request_id: response.request_id }
+      : { ...response, job_id: jobId };
+    ctx.output.result(
+      resultData,
+      ctx.settings.output === "text" ? { text: JSON.stringify(resultData, null, 2) } : undefined,
+    );
+  },
+});
diff --git a/packages/commands/src/commands/cmodel/get.ts b/packages/commands/src/commands/cmodel/get.ts
new file mode 100644
index 00000000..9e8944ff
--- /dev/null
+++ b/packages/commands/src/commands/cmodel/get.ts
@@ -0,0 +1,52 @@
+import { defineCommand, getCustomModelImport } from "bailian-cli-core";
+
+import { IMPORT_JOB_ID_FLAGS, PUBLIC_DASHSCOPE_NOTE, buildImportJobOutput } from "./shared.ts";
+
+export default defineCommand({
+  defaultOutput: "json",
+  description: {
+    "en-US": "Get details of a custom model (import job)",
+    "zh-CN": "获取自定义模型（导入任务）的详情",
+  },
+  auth: "apiKey",
+  usageArgs: "--job-id <id>",
+  flags: IMPORT_JOB_ID_FLAGS,
+  exampleArgs: ["--job-id 937b-xxxx", "--job-id 937b-xxxx --output json"],
+  notes: [
+    PUBLIC_DASHSCOPE_NOTE,
+    {
+      "en-US":
+        "FAILED jobs include the server-provided error_code verbatim; the CLI does not translate it.",
+      "zh-CN": "FAILED 任务会原样返回服务端提供的 error_code，CLI 不做翻译。",
+    },
+  ],
+  async run(ctx) {
+    const { settings, flags } = ctx;
+    const jobId = flags.jobId;
+
+    if (settings.dryRun) {
+      const resultData = { action: "cmodel.get", job_id: jobId };
+      ctx.output.result(
+        resultData,
+        ctx.settings.output === "text" ? { text: JSON.stringify(resultData, null, 2) } : undefined,
+      );
+      return;
+    }
+
+    const response = await getCustomModelImport(ctx.client, jobId);
+    const job = response.output ?? response.data;
+    if (!job) {
+      const resultData = { job_id: jobId, error: "No data returned" };
+      ctx.output.result(
+        resultData,
+        ctx.settings.output === "text" ? { text: JSON.stringify(resultData, null, 2) } : undefined,
+      );
+      return;
+    }
+    const resultData = { ...buildImportJobOutput(job, jobId), request_id: response.request_id };
+    ctx.output.result(
+      resultData,
+      ctx.settings.output === "text" ? { text: JSON.stringify(resultData, null, 2) } : undefined,
+    );
+  },
+});
diff --git a/packages/commands/src/commands/cmodel/import.ts b/packages/commands/src/commands/cmodel/import.ts
new file mode 100644
index 00000000..f9a7b19b
--- /dev/null
+++ b/packages/commands/src/commands/cmodel/import.ts
@@ -0,0 +1,215 @@
+import {
+  defineCommand,
+  createCustomModelImport,
+  findSupportImportModel,
+  listSupportImportModels,
+  requiredTrainTypeForWeightType,
+  UsageError,
+  type CreateCustomModelImportRequest,
+  type FlagsDef,
+} from "bailian-cli-core";
+
+import { PUBLIC_DASHSCOPE_NOTE } from "./shared.ts";
+
+const CREATE_FLAGS = {
+  baseModel: {
+    type: "string",
+    valueHint: "<model>",
+    description: {
+      "en-US": "Base model ID used for tuning (required, e.g. qwen3-32b)",
+      "zh-CN": "训练使用的基础模型 ID（必填，例如 qwen3-32b）",
+    },
+    required: true,
+  },
+  weightType: {
+    type: "string",
+    valueHint: "<type>",
+    choices: ["lora", "full"] as const,
+    description: {
+      "en-US": "Tuning weight type (required): lora or full",
+      "zh-CN": "调优类型（必填）：lora 或 full",
+    },
+    required: true,
+  },
+  bucket: {
+    type: "string",
+    valueHint: "<bucket>",
+    description: {
+      "en-US": "OSS bucket containing the model files (required)",
+      "zh-CN": "存放模型文件的 OSS Bucket（必填）",
+    },
+    required: true,
+  },
+  objectKey: {
+    type: "string",
+    valueHint: "<dir/>",
+    description: {
+      "en-US":
+        "OSS directory prefix containing the model files (required, must end with / and not be the bucket root)",
+      "zh-CN": "模型文件所在 OSS 目录前缀（必填，必须以 / 结尾且不能是 Bucket 根目录）",
+    },
+    required: true,
+  },
+  displayName: {
+    type: "string",
+    valueHint: "<name>",
+    description: {
+      "en-US": "Display name (optional, max 50 characters)",
+      "zh-CN": "显示名称（可选，最多 50 字符）",
+    },
+  },
+} satisfies FlagsDef;
+
+const OSS_BUCKET_PATTERN = /^[a-z0-9][a-z0-9-]{1,61}[a-z0-9]$/;
+
+export default defineCommand({
+  defaultOutput: "json",
+  description: {
+    "en-US": "Import a custom model from OSS (creates an import job)",
+    "zh-CN": "从 OSS 导入自定义模型（创建导入任务）",
+  },
+  auth: "apiKey",
+  optionalConsole: true,
+  usageArgs:
+    "--base-model <model> --weight-type <lora|full> --bucket <bucket> --object-key <dir/> [--display-name <name>]",
+  flags: CREATE_FLAGS,
+  exampleArgs: [
+    "--base-model qwen3-32b --weight-type lora --bucket my-model-bucket --object-key models/qwen3-32b-lora/",
+    "--base-model qwen3-32b --weight-type lora --bucket my-model-bucket --object-key models/qwen3-32b-lora/ --display-name my-lora",
+    "--base-model qwen3-32b --weight-type lora --bucket my-model-bucket --object-key models/qwen3-32b-lora/ --dry-run",
+  ],
+  notes: [
+    PUBLIC_DASHSCOPE_NOTE,
+    {
+      "en-US":
+        "After submission, follow the job with `cmodel watch --job-id` and inspect it with `cmodel get` / `cmodel list`.",
+      "zh-CN":
+        "提交后可用 `cmodel watch --job-id` 跟踪进度，并用 `cmodel get` / `cmodel list` 查看任务。",
+    },
+    {
+      "en-US":
+        "The import API is currently available only in the Beijing region; use the console in other regions.",
+      "zh-CN": "导入 API 当前仅北京地域开放；其他地域请使用控制台操作。",
+    },
+    {
+      "en-US":
+        "The source is fixed to OSS in this release. The bucket must be tagged bailian-datahub-access=read, must not use archive/cold-archive/deep-archive storage or SSE-KMS, and the directory must contain the model files required by the platform (e.g. adapter_model.safetensors, adapter_config.json, config.json).",
+      "zh-CN":
+        "本期导入来源固定为 OSS。Bucket 需添加 bailian-datahub-access=read 标签，不支持归档/冷归档/深度冷归档存储和 SSE-KMS；目录需包含平台要求的模型文件（如 adapter_model.safetensors、adapter_config.json、config.json）。",
+    },
+    {
+      "en-US":
+        "When a valid console session is available, the base model is checked against the console-supported import list before submission: the ID is normalized to the catalog `model` value, and weight-type is checked against the model's supportTrainTypes (lora requires efficient_sft, full requires sft). Unsupported combinations fail immediately with a usage error. If the console session is missing or expired, the check is skipped with a warning and the import API's final result is authoritative.",
+      "zh-CN":
+        "控制台会话可用时，提交前会用控制台支持导入的基础模型清单校验 --base-model：ID 会规范化为目录返回的 model 值，并按模型的 supportTrainTypes 校验 weight-type（lora 需要 efficient_sft，full 需要 sft），不支持的组合会立即报用法错误。控制台登录缺失或过期时跳过校验并输出警告，以导入接口的最终结果为准。",
+    },
+    {
+      "en-US":
+        "The current product version supports importing LoRA models; submitting weight-type=full may be rejected by the service. The API contract keeps both values, so the CLI does not hide `full`.",
+      "zh-CN":
+        "当前产品版本支持导入 LoRA 模型；提交 weight-type=full 可能被服务端拒绝。API 协议保留两个取值，因此 CLI 不隐藏 full。",
+    },
+  ],
+  validate(flags) {
+    if (flags.bucket && !OSS_BUCKET_PATTERN.test(flags.bucket)) {
+      return "--bucket must be a valid OSS bucket name (3-63 lowercase letters, digits or hyphens, starting and ending with a letter or digit). / --bucket 必须是合法的 OSS Bucket 名称（3-63 位小写字母、数字或连字符，且以字母或数字开头和结尾）。";
+    }
+    if (flags.objectKey && (!flags.objectKey.endsWith("/") || flags.objectKey === "/")) {
+      return "--object-key must be a non-root directory prefix ending with /. / --object-key 必须是以 / 结尾的非根目录前缀。";
+    }
+    if (flags.displayName !== undefined && flags.displayName.length > 50) {
+      return "--display-name must be at most 50 characters. / --display-name 最多 50 个字符。";
+    }
+    return undefined;
+  },
+  async run(ctx) {
+    const { settings, flags } = ctx;
+    const buildRequest = (baseModel: string): CreateCustomModelImportRequest => ({
+      model_name: baseModel,
+      ...(flags.displayName ? { display_name: flags.displayName } : {}),
+      source: "oss",
+      weight_type: flags.weightType,
+      storage_info: {
+        bucket_name: flags.bucket,
+        object_key: flags.objectKey,
+      },
+    });
+
+    if (settings.dryRun) {
+      const resultData = {
+        action: "cmodel.import",
+        request: buildRequest(flags.baseModel),
+        precheck: { performed: false, reason: "dry-run" },
+      };
+      ctx.output.result(
+        resultData,
+        ctx.settings.output === "text" ? { text: JSON.stringify(resultData, null, 2) } : undefined,
+      );
+      return;
+    }
+
+    let baseModel = flags.baseModel;
+    let precheck: Record<string, unknown>;
+    try {
+      const supported = await listSupportImportModels(ctx.client);
+      const matched = findSupportImportModel(supported.list, flags.baseModel);
+      if (!matched) {
+        const supportedModels = supported.list
+          .map((entry) => entry.model)
+          .filter((model): model is string => Boolean(model))
+          .sort();
+        throw new UsageError(
+          `Base model "${flags.baseModel}" is not in the current import-supported list. Supported models: ${supportedModels.join(
+            ", ",
+          )}. Run cmodel base-models for the latest list. / 基础模型 "${flags.baseModel}" 不在当前支持导入的模型清单中。支持模型：${supportedModels.join(
+            "、",
+          )}。可运行 cmodel base-models 查看最新清单。`,
+        );
+      }
+      if (matched.model && matched.model !== flags.baseModel) {
+        baseModel = matched.model;
+        ctx.output.info({
+          "en-US": `Base model normalized to ${matched.model} (catalog model ID).`,
+          "zh-CN": `基础模型已规范化为 ${matched.model}（目录返回的模型 ID）。`,
+        });
+      }
+      const supportTrainTypes = matched.supportTrainTypes ?? [];
+      const requiredTrainType = requiredTrainTypeForWeightType(flags.weightType);
+      if (supportTrainTypes.length > 0 && !supportTrainTypes.includes(requiredTrainType)) {
+        const alternativeWeightType = flags.weightType === "lora" ? "full" : "lora";
+        throw new UsageError(
+          `Base model ${baseModel} supports train types: ${supportTrainTypes.join(
+            ", ",
+          )}. weight-type=${flags.weightType} requires ${requiredTrainType}; use --weight-type ${alternativeWeightType}. / 基础模型 ${baseModel} 支持的训练方式：${supportTrainTypes.join(
+            "、",
+          )}。weight-type=${flags.weightType} 需要 ${requiredTrainType}；请改用 --weight-type ${alternativeWeightType}。`,
+        );
+      }
+      precheck = {
+        performed: true,
+        matched: true,
+        ...(baseModel !== flags.baseModel ? { normalized_base_model: baseModel } : {}),
+        support_train_types: supportTrainTypes,
+      };
+    } catch (error) {
+      if (error instanceof UsageError) throw error;
+      precheck = { performed: false, reason: "console-unavailable" };
+      ctx.output.warning({
+        code: "model_import.precheck_skipped",
+        message: {
+          "en-US":
+            "Base-model precheck skipped (console session unavailable or expired). The import API's final result is authoritative.",
+          "zh-CN": "已跳过基础模型预检（控制台会话不可用或已过期）。以导入接口的最终结果为准。",
+        },
+        details: { message: error instanceof Error ? error.message : String(error) },
+      });
+    }
+
+    const response = await createCustomModelImport(ctx.client, buildRequest(baseModel));
+    const resultData = { ...response, precheck };
+    ctx.output.result(
+      resultData,
+      ctx.settings.output === "text" ? { text: JSON.stringify(resultData, null, 2) } : undefined,
+    );
+  },
+});
diff --git a/packages/commands/src/commands/cmodel/list.ts b/packages/commands/src/commands/cmodel/list.ts
new file mode 100644
index 00000000..d4d5f881
--- /dev/null
+++ b/packages/commands/src/commands/cmodel/list.ts
@@ -0,0 +1,122 @@
+import {
+  defineCommand,
+  listCustomModelImports,
+  type CustomModelImportJob,
+  type FlagsDef,
+} from "bailian-cli-core";
+
+import { PUBLIC_DASHSCOPE_NOTE } from "./shared.ts";
+
+const LIST_FLAGS = {
+  page: {
+    type: "number",
+    valueHint: "<n>",
+    description: { "en-US": "Page number (default: 1)", "zh-CN": "页码（默认：1）" },
+  },
+  pageSize: {
+    type: "number",
+    valueHint: "<n>",
+    description: {
+      "en-US": "Results per page (default: 10, max 100)",
+      "zh-CN": "每页结果数（默认：10，最多：100）",
+    },
+  },
+  status: {
+    type: "string",
+    valueHint: "<status>",
+    choices: ["PENDING", "RUNNING", "SUCCESSED", "FAILED"] as const,
+    description: {
+      "en-US": "Filter by status (server-side)",
+      "zh-CN": "按任务状态筛选（服务端筛选）",
+    },
+  },
+  modelName: {
+    type: "string",
+    valueHint: "<name>",
+    description: {
+      "en-US": "Filter by the server-generated model identifier (exact match, server-side)",
+      "zh-CN": "按服务端生成的模型标识精确筛选（服务端筛选）",
+    },
+  },
+} satisfies FlagsDef;
+
+export default defineCommand({
+  defaultOutput: "json",
+  description: {
+    "en-US": "List custom models (import jobs)",
+    "zh-CN": "列出自定义模型（导入任务）",
+  },
+  auth: "apiKey",
+  usageArgs: "[--page <n>] [--page-size <n>] [--status <status>] [--model-name <name>]",
+  flags: LIST_FLAGS,
+  exampleArgs: [
+    "",
+    "--status RUNNING",
+    "--model-name qwen3-32b-offline-20240101-abc1",
+    "--page-size 20",
+  ],
+  notes: [
+    PUBLIC_DASHSCOPE_NOTE,
+    {
+      "en-US":
+        "status and model-name are server-side filters over the whole result set (not local page filters). FAILED entries include the server-provided error_code.",
+      "zh-CN":
+        "status 与 model-name 由服务端在整个结果集上筛选（不是本页本地筛选）。FAILED 条目会包含服务端提供的 error_code。",
+    },
+  ],
+  validate(flags) {
+    if (flags.page !== undefined && (!Number.isFinite(flags.page) || flags.page < 1)) {
+      return "--page must be at least 1. / --page 必须至少为 1。";
+    }
+    if (
+      flags.pageSize !== undefined &&
+      (!Number.isFinite(flags.pageSize) || flags.pageSize < 1 || flags.pageSize > 100)
+    ) {
+      return "--page-size must be between 1 and 100. / --page-size 必须在 1 到 100 之间。";
+    }
+    return undefined;
+  },
+  async run(ctx) {
+    const { settings, flags } = ctx;
+    const params = {
+      pageNo: flags.page,
+      pageSize: flags.pageSize,
+      status: flags.status || undefined,
+      modelName: flags.modelName || undefined,
+    };
+
+    if (settings.dryRun) {
+      const resultData = { action: "cmodel.list", ...params };
+      ctx.output.result(
+        resultData,
+        ctx.settings.output === "text" ? { text: JSON.stringify(resultData, null, 2) } : undefined,
+      );
+      return;
+    }
+
+    const response = await listCustomModelImports(ctx.client, params);
+    const payload = response.output ?? response.data;
+    const jobs: CustomModelImportJob[] = payload?.list ?? [];
+    const items = jobs.map((job) => ({
+      job_id: job.job_id ?? "",
+      model_name: job.model_name ?? "",
+      display_name: job.display_name ?? "",
+      source: job.source ?? "",
+      storage_info: job.storage_info ?? null,
+      status: job.status ?? "",
+      gmt_create: job.gmt_create ?? "",
+      ...(job.error_code !== undefined ? { error_code: job.error_code } : {}),
+    }));
+    const resultData = {
+      items,
+      total: payload?.total ?? items.length,
+      page_no: payload?.page_no ?? params.pageNo ?? 1,
+      page_size: payload?.page_size ?? params.pageSize ?? 10,
+      request_id: response.request_id,
+    };
+    ctx.output.result(
+      resultData,
+      ctx.settings.output === "text" ? { text: JSON.stringify(resultData, null, 2) } : undefined,
+    );
+  },
+});
diff --git a/packages/commands/src/commands/cmodel/shared.ts b/packages/commands/src/commands/cmodel/shared.ts
new file mode 100644
index 00000000..df9f5517
--- /dev/null
+++ b/packages/commands/src/commands/cmodel/shared.ts
@@ -0,0 +1,75 @@
+import { BailianError, ExitCode, type CustomModelImportJob, type FlagsDef } from "bailian-cli-core";
+
+export const IMPORT_JOB_ID_FLAGS = {
+  jobId: {
+    type: "string",
+    valueHint: "<id>",
+    description: {
+      "en-US": "Import job ID returned by `cmodel import` (required)",
+      "zh-CN": "由 `cmodel import` 返回的导入任务 ID（必填）",
+    },
+    required: true,
+  },
+} satisfies FlagsDef;
+
+/** Server-documented terminal statuses (including the `SUCCESSED` spelling). */
+export const IMPORT_TERMINAL_STATUSES = new Set(["SUCCESSED", "FAILED"]);
+
+/** cmodel APIs live on the public DashScope host, not workspace hosts. */
+export const PUBLIC_DASHSCOPE_NOTE = {
+  "en-US":
+    "This command always calls the public DashScope endpoint (https://dashscope.aliyuncs.com) in the Beijing region. The API is not deployed on workspace-specific base URLs (https://<workspace>.cn-beijing.maas.aliyuncs.com, HTTP 404), so the Profile base URL is not used for this request; workspace API keys are accepted.",
+  "zh-CN":
+    "该命令固定调用公共 DashScope 站点（https://dashscope.aliyuncs.com，北京地域）。此 API 未部署在 Workspace 专属 base URL（https://<workspace>.cn-beijing.maas.aliyuncs.com，返回 HTTP 404）上，因此本次请求不使用 Profile 的 base URL；Workspace API Key 可正常调用。",
+};
+
+export function buildImportJobOutput(
+  job: CustomModelImportJob,
+  fallbackJobId: string,
+): Record<string, unknown> {
+  return {
+    job_id: job.job_id ?? fallbackJobId,
+    model_name: job.model_name ?? "",
+    display_name: job.display_name ?? "",
+    source: job.source ?? "",
+    storage_info: job.storage_info ?? null,
+    status: job.status ?? "",
+    gmt_create: job.gmt_create ?? "",
+    ...(job.error_code !== undefined ? { error_code: job.error_code } : {}),
+  };
+}
+
+export function formatImportFailure(
+  jobId: string,
+  status: string,
+  errorCode?: string,
+): BailianError {
+  const detail = errorCode ? ` Server error_code: ${errorCode}` : "";
+  return new BailianError(
+    `Import job ${jobId} ended in status ${status}.${detail} / 导入任务 ${jobId} 已以 ${status} 状态结束。${
+      errorCode ? ` 服务端错误：${errorCode}` : ""
+    }`,
+    ExitCode.GENERAL,
+  );
+}
+
+/**
+ * Resolve after `milliseconds`, rejecting early if `signal` aborts (Ctrl-C).
+ */
+export function sleepWithSignal(milliseconds: number, signal: AbortSignal): Promise<void> {
+  return new Promise((resolve, reject) => {
+    if (signal.aborted) {
+      reject(new Error("aborted"));
+      return;
+    }
+    const onAbort = () => {
+      clearTimeout(timer);
+      reject(new Error("aborted"));
+    };
+    const timer = setTimeout(() => {
+      signal.removeEventListener("abort", onAbort);
+      resolve();
+    }, milliseconds);
+    signal.addEventListener("abort", onAbort, { once: true });
+  });
+}
diff --git a/packages/commands/src/commands/cmodel/watch.ts b/packages/commands/src/commands/cmodel/watch.ts
new file mode 100644
index 00000000..3729b035
--- /dev/null
+++ b/packages/commands/src/commands/cmodel/watch.ts
@@ -0,0 +1,209 @@
+import {
+  BailianError,
+  ExitCode,
+  defineCommand,
+  getCustomModelImport,
+  type FlagsDef,
+} from "bailian-cli-core";
+
+import {
+  IMPORT_JOB_ID_FLAGS,
+  IMPORT_TERMINAL_STATUSES,
+  buildImportJobOutput,
+  formatImportFailure,
+  PUBLIC_DASHSCOPE_NOTE,
+  sleepWithSignal,
+} from "./shared.ts";
+
+const DEFAULT_INTERVAL_SEC = 10;
+const MIN_INTERVAL_SEC = 1;
+
+const WATCH_FLAGS = {
+  ...IMPORT_JOB_ID_FLAGS,
+  follow: {
+    type: "switch",
+    description: {
+      "en-US":
+        "Block and poll until a terminal state. Without it, a single status probe is performed and the command returns immediately.",
+      "zh-CN": "阻塞并轮询至终态。不使用时仅查询一次状态并立即返回。",
+    },
+  },
+  interval: {
+    type: "number",
+    valueHint: "<sec>",
+    description: {
+      "en-US": `Seconds between polls with --follow (default: ${DEFAULT_INTERVAL_SEC}, min: ${MIN_INTERVAL_SEC}). Ignored without --follow.`,
+      "zh-CN": `使用 --follow 时的轮询间隔秒数（默认：${DEFAULT_INTERVAL_SEC}，最小：${MIN_INTERVAL_SEC}）。未使用 --follow 时忽略。`,
+    },
+  },
+  pollTimeout: {
+    type: "number",
+    valueHint: "<sec>",
+    description: {
+      "en-US":
+        "With --follow, stop polling after this many seconds (default: no limit). Ignored without --follow.",
+      "zh-CN": "使用 --follow 时，在指定秒数后停止轮询（默认：无限制）。未使用 --follow 时忽略。",
+    },
+  },
+} satisfies FlagsDef;
+
+function formatElapsed(milliseconds: number): string {
+  const totalSeconds = Math.floor(milliseconds / 1000);
+  const minutes = Math.floor(totalSeconds / 60);
+  const seconds = totalSeconds % 60;
+  if (minutes === 0) return `${seconds}s`;
+  return `${minutes}m ${seconds}s`;
+}
+
+export default defineCommand({
+  defaultOutput: "json",
+  description: {
+    "en-US":
+      "Probe a custom model's import status (default: single non-blocking fetch). Pass --follow to poll until terminal.",
+    "zh-CN": "查询自定义模型的导入状态（默认：单次非阻塞获取）。使用 --follow 持续轮询至终态。",
+  },
+  auth: "apiKey",
+  usageArgs: "--job-id <id> [--follow] [--interval <sec>] [--poll-timeout <sec>]",
+  flags: WATCH_FLAGS,
+  exampleArgs: [
+    "--job-id 937b-xxxx",
+    "--job-id 937b-xxxx --output json",
+    "--job-id 937b-xxxx --follow",
+    "--job-id 937b-xxxx --follow --interval 5 --poll-timeout 3600",
+  ],
+  notes: [
+    PUBLIC_DASHSCOPE_NOTE,
+    {
+      "en-US":
+        "Default (no --follow) is a NON-BLOCKING single status probe for agents/scripts. Terminal statuses are SUCCESSED and FAILED; an unknown server status is surfaced verbatim and treated as non-terminal.",
+      "zh-CN":
+        "默认不使用 --follow，执行非阻塞的单次状态查询，适合 Agent/脚本。终态为 SUCCESSED 和 FAILED；未知服务端状态原样透出并按非终态处理。",
+    },
+    {
+      "en-US":
+        "A terminal FAILED status raises a normal CLI error (non-zero exit) with the server error_code preserved; SUCCEEDED and still-running statuses return 0. With --follow, exceeding --poll-timeout raises a timeout error.",
+      "zh-CN":
+        "终态 FAILED 会触发普通 CLI 错误（非零退出码）并保留服务端 error_code；SUCCEEDED 或仍在运行时返回 0。使用 --follow 时，超过 --poll-timeout 会触发超时错误。",
+    },
+  ],
+  validate(flags) {
+    if (flags.interval !== undefined && (!Number.isFinite(flags.interval) || flags.interval < 1)) {
+      return "--interval must be at least 1 second. / --interval 必须至少为 1 秒。";
+    }
+    if (
+      flags.pollTimeout !== undefined &&
+      (!Number.isFinite(flags.pollTimeout) || flags.pollTimeout < 1)
+    ) {
+      return "--poll-timeout must be at least 1 second. / --poll-timeout 必须至少为 1 秒。";
+    }
+    return undefined;
+  },
+  async run(ctx) {
+    const { settings, flags } = ctx;
+    const jobId = flags.jobId;
+    const follow = flags.follow;
+    const intervalSec = Math.max(MIN_INTERVAL_SEC, flags.interval ?? DEFAULT_INTERVAL_SEC);
+    const pollTimeoutSec = flags.pollTimeout;
+
+    if (settings.dryRun) {
+      const resultData = {
+        action: "cmodel.watch",
+        job_id: jobId,
+        follow,
+        interval: intervalSec,
+        timeout: pollTimeoutSec,
+      };
+      ctx.output.result(
+        resultData,
+        ctx.settings.output === "text" ? { text: JSON.stringify(resultData, null, 2) } : undefined,
+      );
+      return;
+    }
+
+    if (!follow) {
+      const response = await getCustomModelImport(ctx.client, jobId);
+      const job = response.output ?? response.data;
+      const status = String(job?.status ?? "").toUpperCase();
+      const terminal = IMPORT_TERMINAL_STATUSES.has(status);
+      const resultData = job
+        ? { ...buildImportJobOutput(job, jobId), terminal, request_id: response.request_id }
+        : {
+            job_id: jobId,
+            status: "UNKNOWN",
+            terminal,
+            request_id: response.request_id,
+          };
+      ctx.output.result(
+        resultData,
+        ctx.settings.output === "text" ? { text: JSON.stringify(resultData, null, 2) } : undefined,
+      );
+      if (terminal && status !== "SUCCESSED") {
+        throw formatImportFailure(jobId, job?.status ?? status, job?.error_code);
+      }
+      return;
+    }
+
+    const controller = new AbortController();
+    const onSigint = () => controller.abort();
+    process.on("SIGINT", onSigint);
+
+    try {
+      let lastStatus = "";
+      const startedAt = Date.now();
+
+      // eslint-disable-next-line no-constant-condition
+      while (true) {
+        const response = await getCustomModelImport(ctx.client, jobId, controller.signal);
+        const job = response.output ?? response.data;
+        const rawStatus = String(job?.status ?? "");
+        const status = rawStatus.toUpperCase();
+
+        if (status !== lastStatus) {
+          ctx.output.progress({
+            "en-US": `${jobId}  ${rawStatus || "UNKNOWN"}`,
+            "zh-CN": `${jobId}  ${rawStatus || "UNKNOWN"}`,
+          });
+          lastStatus = status;
+        }
+
+        if (IMPORT_TERMINAL_STATUSES.has(status)) {
+          const resultData = {
+            ...(job ? buildImportJobOutput(job, jobId) : { job_id: jobId }),
+            request_id: response.request_id,
+          };
+          ctx.output.result(
+            resultData,
+            ctx.settings.output === "text"
+              ? { text: JSON.stringify(resultData, null, 2) }
+              : undefined,
+          );
+          if (status !== "SUCCESSED") {
+            throw formatImportFailure(jobId, rawStatus, job?.error_code);
+          }
+          return;
+        }
+
+        if (pollTimeoutSec !== undefined && (Date.now() - startedAt) / 1000 >= pollTimeoutSec) {
+          throw new BailianError(
+            `Watching import job ${jobId} timed out after ${formatElapsed(
+              Date.now() - startedAt,
+            )} (last status: ${rawStatus || "UNKNOWN"}). / 跟踪导入任务 ${jobId} 超时（耗时 ${formatElapsed(
+              Date.now() - startedAt,
+            )}，最后状态：${rawStatus || "UNKNOWN"}）。`,
+            ExitCode.TIMEOUT,
+          );
+        }
+
+        await sleepWithSignal(intervalSec * 1000, controller.signal);
+      }
+    } catch (error) {
+      if (controller.signal.aborted) {
+        ctx.output.info({ "en-US": "Interrupted.", "zh-CN": "已中断。" });
+        return;
+      }
+      throw error;
+    } finally {
+      process.off("SIGINT", onSigint);
+    }
+  },
+});
diff --git a/packages/commands/src/index.ts b/packages/commands/src/index.ts
index d9377e67..b06422ab 100644
--- a/packages/commands/src/index.ts
+++ b/packages/commands/src/index.ts
@@ -113,6 +113,12 @@ export { default as advisorRecommend } from "./commands/advisor/recommend.ts";
 export { default as modelList } from "./commands/model/list.ts";
 export { default as modelSearch } from "./commands/model/search.ts";
 export { default as modelCode } from "./commands/model/code.ts";
+export { default as cmodelImport } from "./commands/cmodel/import.ts";
+export { default as cmodelGet } from "./commands/cmodel/get.ts";
+export { default as cmodelList } from "./commands/cmodel/list.ts";
+export { default as cmodelWatch } from "./commands/cmodel/watch.ts";
+export { default as cmodelDelete } from "./commands/cmodel/delete.ts";
+export { default as cmodelBaseModels } from "./commands/cmodel/base-models.ts";
 export { default as workspaceList } from "./commands/workspace/list.ts";
 export { default as quotaList } from "./commands/quota/list.ts";
 export { default as quotaUpdate } from "./commands/quota/update.ts";
diff --git a/packages/commands/tests/e2e/cmodel.e2e.test.ts b/packages/commands/tests/e2e/cmodel.e2e.test.ts
new file mode 100644
index 00000000..1cbdeab8
--- /dev/null
+++ b/packages/commands/tests/e2e/cmodel.e2e.test.ts
@@ -0,0 +1,230 @@
+import { describe, expect, test } from "vite-plus/test";
+import { parseStdoutJson, runCommandHelp, runCommandInProcess } from "./helpers.ts";
+import { CMODEL_ROUTES } from "./topic-routes.ts";
+
+describe("e2e: cmodel", () => {
+  test("cmodel --help 列出子命令", async () => {
+    const { stderr, exitCode } = await runCommandHelp(CMODEL_ROUTES, ["cmodel", "--help"]);
+    expect(exitCode, stderr).toBe(0);
+    expect(stderr).toContain("Usage: bl cmodel <command> [flags]");
+    expect(stderr).toMatch(/import|get|list|watch|delete|base-models/);
+  });
+
+  test("cmodel import --help 展示导入参数与可选 console 预检 flag", async () => {
+    const { stderr, exitCode } = await runCommandHelp(CMODEL_ROUTES, [
+      "cmodel",
+      "import",
+      "--help",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    expect(stderr).toContain("--base-model");
+    expect(stderr).toContain("--weight-type");
+    expect(stderr).toContain("--bucket");
+    expect(stderr).toContain("--object-key");
+    // optionalConsole: apiKey 命令额外可见 console 域 flag
+    expect(stderr).toContain("--console-region");
+    expect(stderr).toContain("API Key");
+  });
+
+  test("cmodel import 缺少 --weight-type 报错", async () => {
+    const { stderr, exitCode } = await runCommandInProcess(CMODEL_ROUTES, [
+      "cmodel",
+      "import",
+      "--base-model",
+      "qwen3-32b",
+      "--bucket",
+      "my-model-bucket",
+      "--object-key",
+      "models/qwen3-32b-lora/",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toContain("--weight-type");
+  });
+
+  test("cmodel import --weight-type 非法取值报错", async () => {
+    const { stderr, exitCode } = await runCommandInProcess(CMODEL_ROUTES, [
+      "cmodel",
+      "import",
+      "--base-model",
+      "qwen3-32b",
+      "--weight-type",
+      "qlora",
+      "--bucket",
+      "my-model-bucket",
+      "--object-key",
+      "models/qwen3-32b-lora/",
+    ]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toContain("--weight-type");
+  });
+
+  test.each([
+    {
+      name: "object-key 不以 / 结尾",
+      objectKey: "models/qwen3-32b-lora",
+      expected: "--object-key",
+    },
+    { name: "object-key 是 Bucket 根目录", objectKey: "/", expected: "--object-key" },
+    {
+      name: "display-name 超过 50 字符",
+      displayName: "x".repeat(51),
+      expected: "--display-name",
+    },
+    { name: "bucket 命名非法", bucket: "My_Model_Bucket", expected: "--bucket" },
+  ])("cmodel 校验失败：$name", async ({ objectKey, displayName, bucket, expected }) => {
+    const args = [
+      "cmodel",
+      "import",
+      "--base-model",
+      "qwen3-32b",
+      "--weight-type",
+      "lora",
+      "--bucket",
+      bucket ?? "my-model-bucket",
+      "--object-key",
+      objectKey ?? "models/qwen3-32b-lora/",
+    ];
+    if (displayName) args.push("--display-name", displayName);
+    const { stderr, exitCode } = await runCommandInProcess(CMODEL_ROUTES, args);
+    expect(exitCode).toBe(2);
+    expect(stderr).toContain(expected);
+  });
+
+  test("cmodel import --dry-run 构造 OSS 导入请求体且跳过预检", async () => {
+    const { stdout, stderr, exitCode } = await runCommandInProcess(CMODEL_ROUTES, [
+      "cmodel",
+      "import",
+      "--base-model",
+      "qwen3-32b",
+      "--weight-type",
+      "lora",
+      "--bucket",
+      "my-model-bucket",
+      "--object-key",
+      "models/qwen3-32b-lora/",
+      "--display-name",
+      "my-lora",
+      "--dry-run",
+      "--output",
+      "json",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    const data = parseStdoutJson<{
+      action?: string;
+      request?: {
+        model_name?: string;
+        display_name?: string;
+        source?: string;
+        weight_type?: string;
+        storage_info?: { bucket_name?: string; object_key?: string };
+      };
+      precheck?: { performed?: boolean; reason?: string };
+    }>(stdout);
+    expect(data.action).toBe("cmodel.import");
+    expect(data.request?.model_name).toBe("qwen3-32b");
+    expect(data.request?.display_name).toBe("my-lora");
+    expect(data.request?.source).toBe("oss");
+    expect(data.request?.weight_type).toBe("lora");
+    expect(data.request?.storage_info).toEqual({
+      bucket_name: "my-model-bucket",
+      object_key: "models/qwen3-32b-lora/",
+    });
+    expect(data.precheck).toEqual({ performed: false, reason: "dry-run" });
+  });
+
+  test.each([
+    {
+      name: "get",
+      args: ["get", "--job-id", "937b-xxxx"],
+      action: "cmodel.get",
+    },
+    {
+      name: "list",
+      args: ["list", "--status", "RUNNING", "--page", "2", "--page-size", "20"],
+      action: "cmodel.list",
+    },
+    {
+      name: "watch",
+      args: ["watch", "--job-id", "937b-xxxx", "--follow", "--interval", "5"],
+      action: "cmodel.watch",
+    },
+    { name: "delete", args: ["delete", "--job-id", "937b-xxxx"], action: "cmodel.delete" },
+  ])("cmodel $name --dry-run 输出计划", async ({ args, action }) => {
+    const { stdout, stderr, exitCode } = await runCommandInProcess(CMODEL_ROUTES, [
+      "cmodel",
+      ...args,
+      "--dry-run",
+      "--output",
+      "json",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    const data = parseStdoutJson<{
+      action?: string;
+      job_id?: string;
+      status?: string;
+      follow?: boolean;
+    }>(stdout);
+    expect(data.action).toBe(action);
+    if (data.job_id) expect(data.job_id).toBe("937b-xxxx");
+    if (action === "cmodel.list") expect(data.status).toBe("RUNNING");
+    if (action === "cmodel.watch") expect(data.follow).toBe(true);
+  });
+
+  test.each([
+    {
+      name: "list --page-size 0",
+      args: ["list", "--page-size", "0"],
+      expected: "--page-size",
+    },
+    {
+      name: "watch --interval 0",
+      args: ["watch", "--job-id", "937b-xxxx", "--interval", "0"],
+      expected: "--interval",
+    },
+  ])("cmodel 校验失败：$name", async ({ args, expected }) => {
+    const { stderr, exitCode } = await runCommandInProcess(CMODEL_ROUTES, ["cmodel", ...args]);
+    expect(exitCode).toBe(2);
+    expect(stderr).toContain(expected);
+  });
+
+  test("cmodel import delete --help 展示高风险提示", async () => {
+    const { stderr, exitCode } = await runCommandHelp(CMODEL_ROUTES, [
+      "cmodel",
+      "delete",
+      "--help",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    expect(stderr).toContain("Risk");
+    expect(stderr).toContain("cannot be undone");
+  });
+
+  test("cmodel import base-models 使用 Console 鉴权并暴露地域 flag", async () => {
+    const { stderr, exitCode } = await runCommandHelp(CMODEL_ROUTES, [
+      "cmodel",
+      "base-models",
+      "--help",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    expect(stderr).toContain("Console");
+    expect(stderr).toContain("--console-region");
+  });
+
+  test("cmodel import base-models --dry-run 输出 console 目标", async () => {
+    const { stdout, stderr, exitCode } = await runCommandInProcess(CMODEL_ROUTES, [
+      "cmodel",
+      "base-models",
+      "--dry-run",
+      "--output",
+      "json",
+    ]);
+    expect(exitCode, stderr).toBe(0);
+    const data = parseStdoutJson<{
+      action?: string;
+      consoleRegion?: string;
+      consoleSite?: string;
+    }>(stdout);
+    expect(data.action).toBe("cmodel.baseModels");
+    expect(data.consoleRegion).toBe("cn-beijing");
+    expect(data.consoleSite).toBe("domestic");
+  });
+});
diff --git a/packages/commands/tests/e2e/topic-routes.ts b/packages/commands/tests/e2e/topic-routes.ts
index f72e4b6a..34a6dbfc 100644
--- a/packages/commands/tests/e2e/topic-routes.ts
+++ b/packages/commands/tests/e2e/topic-routes.ts
@@ -182,6 +182,15 @@ export const MODEL_ROUTES: E2eRouteExports = {
   "model code": "modelCode",
 };
 
+export const CMODEL_ROUTES: E2eRouteExports = {
+  "cmodel import": "cmodelImport",
+  "cmodel get": "cmodelGet",
+  "cmodel list": "cmodelList",
+  "cmodel watch": "cmodelWatch",
+  "cmodel delete": "cmodelDelete",
+  "cmodel base-models": "cmodelBaseModels",
+};
+
 export const QUOTA_ROUTES: E2eRouteExports = {
   "quota list": "quotaList",
   "quota update": "quotaUpdate",
diff --git a/packages/commands/tests/output/managed-command-output.test.ts b/packages/commands/tests/output/managed-command-output.test.ts
index 95a9519d..bf9f403f 100644
--- a/packages/commands/tests/output/managed-command-output.test.ts
+++ b/packages/commands/tests/output/managed-command-output.test.ts
@@ -526,10 +526,20 @@ for (const entry of mapping) {
           else if (nativeStream) {
             expect(result.strayOut).toBe(baseline.strayOut);
             expect(result.stdout).toBe("\n");
-          } else
-            expect(canonical(result.stdout, result.fixtureVersionId)).toBe(
-              canonical(baseline.stdout + baseline.strayOut, baseline.fixtureVersionId),
+          } else {
+            // Version IDs embed a random nonce; when the first 12 characters
+            // happen to match a YAML number pattern (e.g. `0x`), the text
+            // serializer quotes the scalar. Normalize both quote forms.
+            const normalizeScalarQuotes = (value: string) =>
+              value
+                .replaceAll('"<VERSION_ID>"', "<VERSION_ID>")
+                .replaceAll('"<SHORT_VERSION>"', "<SHORT_VERSION>");
+            expect(normalizeScalarQuotes(canonical(result.stdout, result.fixtureVersionId))).toBe(
+              normalizeScalarQuotes(
+                canonical(baseline.stdout + baseline.strayOut, baseline.fixtureVersionId),
+              ),
             );
+          }
           expect(result.calls).toEqual(baseline.calls);
           expect(result.stderr).not.toContain("fixture-api-key");
           expect(result.stderr).not.toContain("fixture-secret-value");
diff --git a/packages/core/package.json b/packages/core/package.json
index 92c65172..eb7f0705 100644
--- a/packages/core/package.json
+++ b/packages/core/package.json
@@ -1,6 +1,6 @@
 {
   "name": "bailian-cli-core",
-  "version": "2.7.2",
+  "version": "2.7.3",
   "description": "Core SDK for bailian-cli. See https://www.npmjs.com/package/bailian-cli for usage.",
   "homepage": "https://bailian.console.aliyun.com/cli",
   "bugs": {
diff --git a/packages/core/src/client/endpoints.ts b/packages/core/src/client/endpoints.ts
index a9acbdeb..c6f80f17 100644
--- a/packages/core/src/client/endpoints.ts
+++ b/packages/core/src/client/endpoints.ts
@@ -314,6 +314,28 @@ export function finetuneCheckpointsPath(jobId: string): string {
   return `/api/v1/fine-tunes/${encodeURIComponent(jobId)}/checkpoints`;
 }
 
+// ---- Custom Model Import (public DashScope host only) ----
+
+/**
+ * Absolute base endpoint for OSS custom-model import jobs.
+ *
+ * This API family is deployed only on the public DashScope host — workspace
+ * hosts (https://<workspace>.cn-beijing.maas.aliyuncs.com) return HTTP 404 for
+ * it, even though they serve other /api/v1 routes. Workspace API keys are
+ * accepted by the public host, so callers keep using the resolved credential.
+ */
+const CUSTOM_MODEL_IMPORT_ENDPOINT = "https://dashscope.aliyuncs.com/api/v1/custom_models/import";
+
+/** Create (POST) and list (GET) endpoint. */
+export function customModelImportEndpoint(): string {
+  return CUSTOM_MODEL_IMPORT_ENDPOINT;
+}
+
+/** Single-import-job get (GET) / delete (DELETE) endpoint. */
+export function customModelImportJobEndpoint(jobId: string): string {
+  return `${CUSTOM_MODEL_IMPORT_ENDPOINT}/${encodeURIComponent(jobId)}`;
+}
+
 /** GET /api/v1/fine-tunes/{job_id}/export/{checkpoint} */
 export function finetuneExportPath(jobId: string, checkpoint: string): string {
   return `/api/v1/fine-tunes/${encodeURIComponent(jobId)}/export/${encodeURIComponent(checkpoint)}`;
diff --git a/packages/core/src/index.ts b/packages/core/src/index.ts
index df16203d..737eb6ef 100644
--- a/packages/core/src/index.ts
+++ b/packages/core/src/index.ts
@@ -16,6 +16,7 @@ export * from "./decision/index.ts";
 export * from "./parsex/index.ts";
 export * from "./finetune/index.ts";
 export * from "./deploy/index.ts";
+export * from "./model-import/index.ts";
 export * from "./speech/index.ts";
 export * from "./types/index.ts";
 export * from "./utils/index.ts";
diff --git a/packages/core/src/model-import/api.ts b/packages/core/src/model-import/api.ts
new file mode 100644
index 00000000..612406bb
--- /dev/null
+++ b/packages/core/src/model-import/api.ts
@@ -0,0 +1,84 @@
+/**
+ * Custom model import HTTP API wrappers.
+ *
+ * Thin functions over `requestJson` for DashScope
+ * `/api/v1/custom_models/import`. Server errors are surfaced verbatim by the
+ * shared HTTP error layer — these wrappers never translate them.
+ */
+import { customModelImportEndpoint, customModelImportJobEndpoint } from "../client/endpoints.ts";
+import type { Client } from "../client/client.ts";
+import type {
+  CreateCustomModelImportRequest,
+  CreateCustomModelImportResponse,
+  DeleteCustomModelImportResponse,
+  GetCustomModelImportResponse,
+  ListCustomModelImportsResponse,
+} from "./types.ts";
+
+/** POST /api/v1/custom_models/import */
+export async function createCustomModelImport(
+  client: Client,
+  body: CreateCustomModelImportRequest,
+  signal?: AbortSignal,
+): Promise<CreateCustomModelImportResponse> {
+  return client.requestJson<CreateCustomModelImportResponse>({
+    path: customModelImportEndpoint(),
+    method: "POST",
+    body,
+    signal,
+  });
+}
+
+export interface ListCustomModelImportsParams {
+  pageNo?: number;
+  pageSize?: number;
+  status?: string;
+  /** Exact match against the server-generated `model_name`. */
+  modelName?: string;
+  signal?: AbortSignal;
+}
+
+/** GET /api/v1/custom_models/import */
+export async function listCustomModelImports(
+  client: Client,
+  params: ListCustomModelImportsParams = {},
+): Promise<ListCustomModelImportsResponse> {
+  const query = new URLSearchParams();
+  if (params.pageNo !== undefined) query.set("page_no", String(params.pageNo));
+  if (params.pageSize !== undefined) query.set("page_size", String(params.pageSize));
+  if (params.status) query.set("status", params.status);
+  if (params.modelName) query.set("model_name", params.modelName);
+  const basePath = customModelImportEndpoint();
+  const path = query.toString() ? `${basePath}?${query.toString()}` : basePath;
+  return client.requestJson<ListCustomModelImportsResponse>({
+    path,
+    method: "GET",
+    signal: params.signal,
+  });
+}
+
+/** GET /api/v1/custom_models/import/{job_id} */
+export async function getCustomModelImport(
+  client: Client,
+  jobId: string,
+  signal?: AbortSignal,
+): Promise<GetCustomModelImportResponse> {
+  return client.requestJson<GetCustomModelImportResponse>({
+    path: customModelImportJobEndpoint(jobId),
+    method: "GET",
+    signal,
+  });
+}
+
+/** DELETE /api/v1/custom_models/import/{job_id} */
+export async function deleteCustomModelImport(
+  client: Client,
+  jobId: string,
+  signal?: AbortSignal,
+): Promise<DeleteCustomModelImportResponse> {
+  return client.requestJson<DeleteCustomModelImportResponse>({
+    path: customModelImportJobEndpoint(jobId),
+    method: "DELETE",
+    signal,
+  });
+}
diff --git a/packages/core/src/model-import/console.ts b/packages/core/src/model-import/console.ts
new file mode 100644
index 00000000..0c89baad
--- /dev/null
+++ b/packages/core/src/model-import/console.ts
@@ -0,0 +1,58 @@
+/**
+ * Console-domain import catalog: base models that support OSS import.
+ *
+ * The frontend page calls the same zelda API for its "base model" picker.
+ * The gateway layer injects `cornerstoneParam` (protocol/console/productCode/
+ * switchUserType/consoleSite) and the console access token; the business data
+ * payload is empty.
+ */
+import type { Client } from "../client/client.ts";
+import { unwrapResponse } from "../console/models.ts";
+import type { SupportImportModel, SupportImportModelsResult } from "./types.ts";
+
+export const LIST_SUPPORT_IMPORT_MODELS_API =
+  "zeldaEasy.broadscope-platform.customModel.listSupportImportModels";
+
+/**
+ * List base models that currently support custom-model import.
+ *
+ * Requires a console credential (`client.console` throws AUTH without one);
+ * callers that treat this as an optional precheck must catch that error.
+ */
+export async function listSupportImportModels(client: Client): Promise<SupportImportModelsResult> {
+  const raw = await client.console<Record<string, unknown>>(LIST_SUPPORT_IMPORT_MODELS_API, {});
+  const response = unwrapResponse(raw) as {
+    total?: number;
+    pageNo?: number;
+    pageSize?: number;
+    list?: SupportImportModel[];
+  };
+  return {
+    total: response.total,
+    pageNo: response.pageNo,
+    pageSize: response.pageSize,
+    list: response.list ?? [],
+  };
+}
+
+/** Exact (case-insensitive) match on the catalog `model` ID. */
+export function findSupportImportModel(
+  models: readonly SupportImportModel[],
+  baseModel: string,
+): SupportImportModel | undefined {
+  const target = baseModel.trim().toLowerCase();
+  return models.find((entry) => String(entry.model ?? "").toLowerCase() === target);
+}
+
+/**
+ * Console train type required by an import weight type.
+ *
+ * The catalog's `supportTrainTypes` uses the fine-tune vocabulary: `sft`
+ * (full-parameter) and `efficient_sft` (LoRA). Mapping the import
+ * `weight_type` onto that vocabulary lets callers reject unsupported
+ * combinations before the service answers with its opaque
+ * "No import access for base model" error.
+ */
+export function requiredTrainTypeForWeightType(weightType: string): "sft" | "efficient_sft" {
+  return weightType === "lora" ? "efficient_sft" : "sft";
+}
diff --git a/packages/core/src/model-import/index.ts b/packages/core/src/model-import/index.ts
new file mode 100644
index 00000000..1ffa9d38
--- /dev/null
+++ b/packages/core/src/model-import/index.ts
@@ -0,0 +1,3 @@
+export * from "./types.ts";
+export * from "./api.ts";
+export * from "./console.ts";
diff --git a/packages/core/src/model-import/types.ts b/packages/core/src/model-import/types.ts
new file mode 100644
index 00000000..514a4f2c
--- /dev/null
+++ b/packages/core/src/model-import/types.ts
@@ -0,0 +1,106 @@
+/**
+ * Custom model import API types.
+ *
+ * Maps DashScope `/api/v1/custom_models/import` request/response shapes
+ * (snake_case preserved verbatim — callers decide how to surface fields).
+ */
+
+/** OSS storage location of the model files to import. */
+export interface CustomModelImportStorageInfo {
+  bucket_name: string;
+  /** Directory prefix; must end with `/` and cannot be the bucket root. */
+  object_key: string;
+}
+
+/** POST /api/v1/custom_models/import request body. */
+export interface CreateCustomModelImportRequest {
+  /** Base model ID (e.g. `qwen3-32b`). */
+  model_name: string;
+  /** Display name; server defaults to the base model name when omitted. */
+  display_name?: string;
+  /** Import source. Only `oss` is currently supported. */
+  source: string;
+  /** `lora` (LoRA-tuned) or `full` (fully-tuned). */
+  weight_type: string;
+  storage_info: CustomModelImportStorageInfo;
+}
+
+/** A custom model import job as returned by the platform. */
+export interface CustomModelImportJob {
+  job_id?: string;
+  /** Server-generated model identifier (base model name + timestamp suffix). */
+  model_name?: string;
+  display_name?: string;
+  /** Uppercase `OSS` in responses. */
+  source?: string;
+  /** Present on create; omitted by get/list. */
+  weight_type?: string;
+  storage_info?: CustomModelImportStorageInfo;
+  /** PENDING | RUNNING | SUCCESSED | FAILED (server values verbatim). */
+  status?: string;
+  /** Present when status is FAILED; server-provided reason text. */
+  error_code?: string;
+  gmt_create?: string;
+  [key: string]: unknown;
+}
+
+export interface CreateCustomModelImportResponse {
+  request_id?: string;
+  output?: CustomModelImportJob;
+  data?: CustomModelImportJob;
+}
+
+export interface GetCustomModelImportResponse {
+  request_id?: string;
+  output?: CustomModelImportJob;
+  data?: CustomModelImportJob;
+}
+
+export interface DeleteCustomModelImportResponse {
+  request_id?: string;
+  output?: CustomModelImportJob;
+  data?: CustomModelImportJob;
+}
+
+export interface ListCustomModelImportsResponse {
+  request_id?: string;
+  output?: {
+    total?: number;
+    page_no?: number;
+    page_size?: number;
+    list?: CustomModelImportJob[];
+    [key: string]: unknown;
+  };
+  data?: {
+    total?: number;
+    page_no?: number;
+    page_size?: number;
+    list?: CustomModelImportJob[];
+    [key: string]: unknown;
+  };
+}
+
+/** Entry returned by the console `listSupportImportModels` catalog API. */
+export interface SupportImportModel {
+  /** Display name (e.g. `Qwen3-32B`); may be absent. */
+  name?: string;
+  /** Model ID used to match `--base-model` (e.g. `qwen3-32b`). */
+  model?: string;
+  /** Training methods supported for import (e.g. `sft`, `efficient_sft`). */
+  supportTrainTypes?: string[];
+  postProcessCapabilities?: {
+    quantizationSupported?: boolean;
+    mtpSupported?: boolean;
+    quantizationStrategyCodes?: string[];
+    [key: string]: unknown;
+  };
+  [key: string]: unknown;
+}
+
+/** Parsed `listSupportImportModels` catalog result. */
+export interface SupportImportModelsResult {
+  total?: number;
+  pageNo?: number;
+  pageSize?: number;
+  list: SupportImportModel[];
+}
diff --git a/packages/core/src/types/command.ts b/packages/core/src/types/command.ts
index 3d7e6822..e5e6c131 100644
--- a/packages/core/src/types/command.ts
+++ b/packages/core/src/types/command.ts
@@ -230,8 +230,13 @@ export type SourceFlags = ParsedFlags<
 >;
 
 /** 该命令可见的凭证域 flag 定义。 */
-export function credentialFlagDefs(cmd: { auth: AuthRequirement }): FlagsDef {
-  if (cmd.auth === "apiKey") return MODEL_AUTH_FLAGS;
+export function credentialFlagDefs(cmd: {
+  auth: AuthRequirement;
+  optionalConsole?: boolean;
+}): FlagsDef {
+  if (cmd.auth === "apiKey") {
+    return cmd.optionalConsole ? { ...MODEL_AUTH_FLAGS, ...CONSOLE_AUTH_FLAGS } : MODEL_AUTH_FLAGS;
+  }
   if (cmd.auth === "console") return CONSOLE_AUTH_FLAGS;
   if (cmd.auth === "openapi") return OPENAPI_AUTH_FLAGS;
   return {};
@@ -313,6 +318,14 @@ export interface Command<F extends FlagsDef = FlagsDef, Prepared = undefined> {
   description: LocalizedText;
   /** Credential this command requires. See {@link AuthRequirement}. */
   auth: AuthRequirement;
+  /**
+   * Best-effort secondary console credential for `auth: "apiKey"` commands.
+   * When true, the runtime additionally resolves the console credential and
+   * injects it when available; a missing/expired console session never blocks
+   * the command. Console-domain flags become visible so the precheck target can
+   * be selected explicitly.
+   */
+  optionalConsole?: boolean;
   /**
    * Runtime-classified operation risk and its user-facing consequence message.
    * Omit for normal commands.
diff --git a/packages/core/tests/model-import-api.test.ts b/packages/core/tests/model-import-api.test.ts
new file mode 100644
index 00000000..dfbc4269
--- /dev/null
+++ b/packages/core/tests/model-import-api.test.ts
@@ -0,0 +1,108 @@
+import { describe, expect, test, vi } from "vite-plus/test";
+import {
+  createCustomModelImport,
+  deleteCustomModelImport,
+  findSupportImportModel,
+  getCustomModelImport,
+  listCustomModelImports,
+  listSupportImportModels,
+  LIST_SUPPORT_IMPORT_MODELS_API,
+  requiredTrainTypeForWeightType,
+  type Client,
+} from "../src/index.ts";
+
+function mockClient(response: unknown = {}) {
+  const requestJson = vi.fn<Client["requestJson"]>().mockResolvedValue(response);
+  return { client: { requestJson } as unknown as Client, requestJson };
+}
+
+describe("custom model import API", () => {
+  test("maps create, get, list, and delete endpoints", async () => {
+    const { client, requestJson } = mockClient();
+    await createCustomModelImport(client, {
+      model_name: "qwen3-32b",
+      source: "oss",
+      weight_type: "lora",
+      storage_info: { bucket_name: "my-model-bucket", object_key: "models/qwen3-32b-lora/" },
+    });
+    await getCustomModelImport(client, "job /?#");
+    await listCustomModelImports(client, {
+      pageNo: 2,
+      pageSize: 20,
+      status: "RUNNING",
+      modelName: "qwen3-32b-offline-20240101-abc1",
+    });
+    await deleteCustomModelImport(client, "job /?#");
+
+    expect(requestJson.mock.calls.map(([options]) => [options.method, options.path])).toEqual([
+      ["POST", "https://dashscope.aliyuncs.com/api/v1/custom_models/import"],
+      ["GET", "https://dashscope.aliyuncs.com/api/v1/custom_models/import/job%20%2F%3F%23"],
+      [
+        "GET",
+        "https://dashscope.aliyuncs.com/api/v1/custom_models/import?page_no=2&page_size=20&status=RUNNING&model_name=qwen3-32b-offline-20240101-abc1",
+      ],
+      ["DELETE", "https://dashscope.aliyuncs.com/api/v1/custom_models/import/job%20%2F%3F%23"],
+    ]);
+    expect(requestJson.mock.calls[0]?.[0].body).toEqual({
+      model_name: "qwen3-32b",
+      source: "oss",
+      weight_type: "lora",
+      storage_info: { bucket_name: "my-model-bucket", object_key: "models/qwen3-32b-lora/" },
+    });
+  });
+
+  test("keeps list query empty when no filters are provided", async () => {
+    const { client, requestJson } = mockClient();
+    await listCustomModelImports(client);
+    expect(requestJson).toHaveBeenCalledWith({
+      path: "https://dashscope.aliyuncs.com/api/v1/custom_models/import",
+      method: "GET",
+      signal: undefined,
+    });
+  });
+});
+
+describe("console-supported import model catalog", () => {
+  test("calls listSupportImportModels with an empty business payload and unwraps DataV2", async () => {
+    const consoleCall = vi
+      .fn<Client["console"]>()
+      .mockResolvedValue(gatewayResponse([{ model: "qwen3-32b", supportTrainTypes: ["sft"] }]));
+    const client = { console: consoleCall } as unknown as Client;
+
+    const result = await listSupportImportModels(client);
+
+    expect(consoleCall).toHaveBeenCalledWith(LIST_SUPPORT_IMPORT_MODELS_API, {});
+    expect(result.total).toBe(1);
+    expect(result.list).toEqual([{ model: "qwen3-32b", supportTrainTypes: ["sft"] }]);
+  });
+
+  test("matches the catalog model ID case-insensitively", () => {
+    const models = [{ model: "qwen3-32b" }, { model: "deepseek-v4-flash-0731" }];
+    expect(findSupportImportModel(models, "Qwen3-32B")?.model).toBe("qwen3-32b");
+    expect(findSupportImportModel(models, "qwen3-8b")).toBeUndefined();
+  });
+
+  test("maps import weight types onto console train types", () => {
+    expect(requiredTrainTypeForWeightType("lora")).toBe("efficient_sft");
+    expect(requiredTrainTypeForWeightType("full")).toBe("sft");
+  });
+});
+
+/** Mirrors the real console gateway envelope (DataV2 double wrap). */
+function gatewayResponse(list: unknown[]) {
+  return {
+    code: "200",
+    data: {
+      DataV2: {
+        ret: ["SUCCESS::接口调用成功"],
+        data: {
+          code: "200",
+          data: { total: list.length, pageNo: 1, pageSize: list.length, list },
+          success: true,
+          appName: "broadscope",
+        },
+      },
+      success: true,
+    },
+  };
+}
diff --git a/packages/kscli/package.json b/packages/kscli/package.json
index ccc9c27a..4adc00d4 100644
--- a/packages/kscli/package.json
+++ b/packages/kscli/package.json
@@ -1,6 +1,6 @@
 {
   "name": "knowledge-studio-cli",
-  "version": "2.7.2",
+  "version": "2.7.3",
   "description": "Lightweight RAG CLI for Aliyun Model Studio — focused on knowledge-base retrieval.",
   "keywords": [
     "alibaba-cloud",
diff --git a/packages/runtime/package.json b/packages/runtime/package.json
index dfbe02ec..57f08c9e 100644
--- a/packages/runtime/package.json
+++ b/packages/runtime/package.json
@@ -1,6 +1,6 @@
 {
   "name": "bailian-cli-runtime",
-  "version": "2.7.2",
+  "version": "2.7.3",
   "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": {
diff --git a/packages/runtime/src/middleware.ts b/packages/runtime/src/middleware.ts
index 0e34e15d..61027fce 100644
--- a/packages/runtime/src/middleware.ts
+++ b/packages/runtime/src/middleware.ts
@@ -216,6 +216,23 @@ export const authStage: Middleware = async (ctx, next) => {
       ...resolveModelBaseUrlState(apiSources),
       apiCred: cred,
     });
+    // Optional console precheck surface: inject the console credential when it
+    // resolves, but never fail an apiKey command because the console session is
+    // missing or expired. Commands must treat `client.console` calls as
+    // best-effort and catch AUTH errors themselves.
+    if (command.optionalConsole) {
+      try {
+        const consoleCred = resolveConsole(sources);
+        ctx.client = new Client({
+          ...base,
+          ...resolveModelBaseUrlState(apiSources),
+          apiCred: cred,
+          consoleCred,
+        });
+      } catch {
+        // Best-effort only — the command's apiKey flow continues.
+      }
+    }
     if (cred && !command.prepare) maybeShowStatusBar(settings, cred.token, cred, ctx.output);
   } else if (command.auth === "console") {
     let cred: ConsoleCredential | undefined;
diff --git a/packages/runtime/tests/console-flags.e2e.test.ts b/packages/runtime/tests/console-flags.e2e.test.ts
index dd022f5b..cccdfcd8 100644
--- a/packages/runtime/tests/console-flags.e2e.test.ts
+++ b/packages/runtime/tests/console-flags.e2e.test.ts
@@ -50,5 +50,15 @@ describe.each([
       expect(stdout).toBe("");
       expect(stderr).toMatch(new RegExp(`${unknownFlag}.*--api-key`));
     });
+
+    test("optionalConsole 命令接受 --console-region 且 dry-run 正常退出", async () => {
+      const { stdout, stderr, exitCode } = await runRuntimeHarness(
+        ["cmodel", "import", "--console-region", "cn-beijing", "--dry-run"],
+        language,
+      );
+      expect(exitCode, stderr).toBe(0);
+      expect(stderr).toBe("");
+      expect(stdout).toBe("");
+    });
   },
 );
diff --git a/packages/runtime/tests/harness/cross-domain-main.ts b/packages/runtime/tests/harness/cross-domain-main.ts
index c585486d..b8f710b2 100644
--- a/packages/runtime/tests/harness/cross-domain-main.ts
+++ b/packages/runtime/tests/harness/cross-domain-main.ts
@@ -22,8 +22,17 @@ const mcpList = defineCommand({
   run: noop,
 });
 
+/** apiKey + optionalConsole stub：用于可选 console 预检 flag 可见性测试 */
+const cmodelImport = defineCommand({
+  description: "runtime e2e stub",
+  auth: "apiKey",
+  optionalConsole: true,
+  flags: {},
+  run: noop,
+});
+
 void createCli(
-  { "text chat": textChat, "mcp list": mcpList },
+  { "text chat": textChat, "mcp list": mcpList, "cmodel import": cmodelImport },
   {
     binName: "bl",
     version: pkg.version,
diff --git a/skills/bailian-cli/SKILL.md b/skills/bailian-cli/SKILL.md
index 8af88875..1c0266ed 100644
--- a/skills/bailian-cli/SKILL.md
+++ b/skills/bailian-cli/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: bailian-cli
 metadata:
-  version: "2.7.2"
+  version: "2.7.3"
   requires:
     bins: ["bl"]
 description: >-
diff --git a/skills/bailian-cli/reference/cmodel.md b/skills/bailian-cli/reference/cmodel.md
new file mode 100644
index 00000000..63ab162e
--- /dev/null
+++ b/skills/bailian-cli/reference/cmodel.md
@@ -0,0 +1,265 @@
+# `bl cmodel` commands
+
+> Auto-generated from `packages/cli/src/commands.ts`. Do not edit by hand.
+> Regenerate: `pnpm --filter bailian-cli run generate:reference`.
+
+Index: [index.md](index.md)
+
+## Commands in this group
+
+| Command                 | Authentication | Description                                                                                                      |
+| ----------------------- | -------------- | ---------------------------------------------------------------------------------------------------------------- |
+| `bl cmodel base-models` | Console        | List base models that can be imported as custom models                                                           |
+| `bl cmodel delete`      | API Key        | Delete a custom model and its Bailian-side import record                                                         |
+| `bl cmodel get`         | API Key        | Get details of a custom model (import job)                                                                       |
+| `bl cmodel import`      | API Key        | Import a custom model from OSS (creates an import job)                                                           |
+| `bl cmodel list`        | API Key        | List custom models (import jobs)                                                                                 |
+| `bl cmodel watch`       | API Key        | Probe a custom model's import status (default: single non-blocking fetch). Pass --follow to poll until terminal. |
+
+## Command details
+
+### `bl cmodel base-models`
+
+| Field              | Value                                                  |
+| ------------------ | ------------------------------------------------------ |
+| **Name**           | `cmodel base-models`                                   |
+| **Description**    | List base models that can be imported as custom models |
+| **Authentication** | Console                                                |
+| **Usage**          | `bl cmodel base-models`                                |
+
+#### Flags
+
+| Flag                           | Type   | Required | Description                                              |
+| ------------------------------ | ------ | -------- | -------------------------------------------------------- |
+| `--console-region <region>`    | string | no       | Console gateway region (e.g. cn-beijing, ap-southeast-1) |
+| `--console-site <site>`        | string | no       | Console site: domestic, international                    |
+| `--console-switch-agent <uid>` | number | no       | Switch agent UID for delegated access                    |
+| `--workspace-id <id>`          | string | no       | Workspace ID (env: BAILIAN_WORKSPACE_ID)                 |
+
+#### Notes
+
+- The list is returned dynamically by the console API and may change between releases. Import is currently available only in the Beijing region.
+
+#### Examples
+
+```bash
+bl cmodel base-models
+```
+
+```bash
+bl cmodel base-models --console-region cn-beijing
+```
+
+```bash
+bl cmodel base-models --output json
+```
+
+### `bl cmodel delete`
+
+| Field              | Value                                                                                                                                    |
+| ------------------ | ---------------------------------------------------------------------------------------------------------------------------------------- |
+| **Name**           | `cmodel delete`                                                                                                                          |
+| **Description**    | Delete a custom model and its Bailian-side import record                                                                                 |
+| **Authentication** | API Key                                                                                                                                  |
+| **Usage**          | `bl cmodel delete --job-id <id>`                                                                                                         |
+| **Risk**           | `high`                                                                                                                                   |
+| **Risk message**   | This permanently deletes the Bailian-side import record and imported model copy; it cannot be undone. OSS source files are not modified. |
+
+> **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                                          |
+| ------------------ | ------ | -------- | ---------------------------------------------------- |
+| `--job-id <id>`    | string | yes      | Import job ID returned by `cmodel import` (required) |
+| `--yes`            | switch | no       | Confirm this high-risk operation                     |
+| `--api-key <key>`  | string | no       | API key                                              |
+| `--base-url <url>` | string | no       | API base URL                                         |
+
+#### Notes
+
+- This command always calls the public DashScope endpoint (https://dashscope.aliyuncs.com) in the Beijing region. The API is not deployed on workspace-specific base URLs (https://<workspace>.cn-beijing.maas.aliyuncs.com, HTTP 404), so the Profile base URL is not used for this request; workspace API keys are accepted.
+- Irreversible: the import record and imported model are removed from Bailian and must be imported again to restore. OSS source files are untouched.
+- Only SUCCESSED or FAILED jobs can be deleted; deleting a RUNNING job is rejected by the service (OperationDenied) and the error is surfaced verbatim.
+
+#### Examples
+
+```bash
+bl cmodel delete --job-id 937b-xxxx
+```
+
+```bash
+bl cmodel delete --job-id 937b-xxxx --dry-run
+```
+
+```bash
+# Only after explicit user confirmation:
+bl cmodel delete --job-id 937b-xxxx --yes
+```
+
+### `bl cmodel get`
+
+| Field              | Value                                      |
+| ------------------ | ------------------------------------------ |
+| **Name**           | `cmodel get`                               |
+| **Description**    | Get details of a custom model (import job) |
+| **Authentication** | API Key                                    |
+| **Usage**          | `bl cmodel get --job-id <id>`              |
+
+#### Flags
+
+| Flag               | Type   | Required | Description                                          |
+| ------------------ | ------ | -------- | ---------------------------------------------------- |
+| `--job-id <id>`    | string | yes      | Import job ID returned by `cmodel import` (required) |
+| `--api-key <key>`  | string | no       | API key                                              |
+| `--base-url <url>` | string | no       | API base URL                                         |
+
+#### Notes
+
+- This command always calls the public DashScope endpoint (https://dashscope.aliyuncs.com) in the Beijing region. The API is not deployed on workspace-specific base URLs (https://<workspace>.cn-beijing.maas.aliyuncs.com, HTTP 404), so the Profile base URL is not used for this request; workspace API keys are accepted.
+- FAILED jobs include the server-provided error_code verbatim; the CLI does not translate it.
+
+#### Examples
+
+```bash
+bl cmodel get --job-id 937b-xxxx
+```
+
+```bash
+bl cmodel get --job-id 937b-xxxx --output json
+```
+
+### `bl cmodel import`
+
+| Field              | Value                                                                                                                            |
+| ------------------ | -------------------------------------------------------------------------------------------------------------------------------- |
+| **Name**           | `cmodel import`                                                                                                                  |
+| **Description**    | Import a custom model from OSS (creates an import job)                                                                           |
+| **Authentication** | API Key                                                                                                                          |
+| **Usage**          | `bl cmodel import --base-model <model> --weight-type <lora\|full> --bucket <bucket> --object-key <dir/> [--display-name <name>]` |
+
+#### Flags
+
+| Flag                           | Type   | Required | Description                                                                                            |
+| ------------------------------ | ------ | -------- | ------------------------------------------------------------------------------------------------------ |
+| `--base-model <model>`         | string | yes      | Base model ID used for tuning (required, e.g. qwen3-32b)                                               |
+| `--weight-type <lora\|full>`   | string | yes      | Tuning weight type (required): lora or full                                                            |
+| `--bucket <bucket>`            | string | yes      | OSS bucket containing the model files (required)                                                       |
+| `--object-key <dir/>`          | string | yes      | OSS directory prefix containing the model files (required, must end with / and not be the bucket root) |
+| `--display-name <name>`        | string | no       | Display name (optional, max 50 characters)                                                             |
+| `--api-key <key>`              | string | no       | API key                                                                                                |
+| `--base-url <url>`             | string | no       | API base URL                                                                                           |
+| `--console-region <region>`    | string | no       | Console gateway region (e.g. cn-beijing, ap-southeast-1)                                               |
+| `--console-site <site>`        | string | no       | Console site: domestic, international                                                                  |
+| `--console-switch-agent <uid>` | number | no       | Switch agent UID for delegated access                                                                  |
+| `--workspace-id <id>`          | string | no       | Workspace ID (env: BAILIAN_WORKSPACE_ID)                                                               |
+
+#### Notes
+
+- This command always calls the public DashScope endpoint (https://dashscope.aliyuncs.com) in the Beijing region. The API is not deployed on workspace-specific base URLs (https://<workspace>.cn-beijing.maas.aliyuncs.com, HTTP 404), so the Profile base URL is not used for this request; workspace API keys are accepted.
+- After submission, follow the job with `cmodel watch --job-id` and inspect it with `cmodel get` / `cmodel list`.
+- The import API is currently available only in the Beijing region; use the console in other regions.
+- The source is fixed to OSS in this release. The bucket must be tagged bailian-datahub-access=read, must not use archive/cold-archive/deep-archive storage or SSE-KMS, and the directory must contain the model files required by the platform (e.g. adapter_model.safetensors, adapter_config.json, config.json).
+- When a valid console session is available, the base model is checked against the console-supported import list before submission: the ID is normalized to the catalog `model` value, and weight-type is checked against the model's supportTrainTypes (lora requires efficient_sft, full requires sft). Unsupported combinations fail immediately with a usage error. If the console session is missing or expired, the check is skipped with a warning and the import API's final result is authoritative.
+- The current product version supports importing LoRA models; submitting weight-type=full may be rejected by the service. The API contract keeps both values, so the CLI does not hide `full`.
+
+#### Examples
+
+```bash
+bl cmodel import --base-model qwen3-32b --weight-type lora --bucket my-model-bucket --object-key models/qwen3-32b-lora/
+```
+
+```bash
+bl cmodel import --base-model qwen3-32b --weight-type lora --bucket my-model-bucket --object-key models/qwen3-32b-lora/ --display-name my-lora
+```
+
+```bash
+bl cmodel import --base-model qwen3-32b --weight-type lora --bucket my-model-bucket --object-key models/qwen3-32b-lora/ --dry-run
+```
+
+### `bl cmodel list`
+
+| Field              | Value                                                                                     |
+| ------------------ | ----------------------------------------------------------------------------------------- |
+| **Name**           | `cmodel list`                                                                             |
+| **Description**    | List custom models (import jobs)                                                          |
+| **Authentication** | API Key                                                                                   |
+| **Usage**          | `bl cmodel list [--page <n>] [--page-size <n>] [--status <status>] [--model-name <name>]` |
+
+#### Flags
+
+| Flag                                             | Type   | Required | Description                                                                |
+| ------------------------------------------------ | ------ | -------- | -------------------------------------------------------------------------- |
+| `--page <n>`                                     | number | no       | Page number (default: 1)                                                   |
+| `--page-size <n>`                                | number | no       | Results per page (default: 10, max 100)                                    |
+| `--status <PENDING\|RUNNING\|SUCCESSED\|FAILED>` | string | no       | Filter by status (server-side)                                             |
+| `--model-name <name>`                            | string | no       | Filter by the server-generated model identifier (exact match, server-side) |
+| `--api-key <key>`                                | string | no       | API key                                                                    |
+| `--base-url <url>`                               | string | no       | API base URL                                                               |
+
+#### Notes
+
+- This command always calls the public DashScope endpoint (https://dashscope.aliyuncs.com) in the Beijing region. The API is not deployed on workspace-specific base URLs (https://<workspace>.cn-beijing.maas.aliyuncs.com, HTTP 404), so the Profile base URL is not used for this request; workspace API keys are accepted.
+- status and model-name are server-side filters over the whole result set (not local page filters). FAILED entries include the server-provided error_code.
+
+#### Examples
+
+```bash
+bl cmodel list
+```
+
+```bash
+bl cmodel list --status RUNNING
+```
+
+```bash
+bl cmodel list --model-name qwen3-32b-offline-20240101-abc1
+```
+
+```bash
+bl cmodel list --page-size 20
+```
+
+### `bl cmodel watch`
+
+| Field              | Value                                                                                                            |
+| ------------------ | ---------------------------------------------------------------------------------------------------------------- |
+| **Name**           | `cmodel watch`                                                                                                   |
+| **Description**    | Probe a custom model's import status (default: single non-blocking fetch). Pass --follow to poll until terminal. |
+| **Authentication** | API Key                                                                                                          |
+| **Usage**          | `bl cmodel watch --job-id <id> [--follow] [--interval <sec>] [--poll-timeout <sec>]`                             |
+
+#### Flags
+
+| Flag                   | Type   | Required | Description                                                                                                                |
+| ---------------------- | ------ | -------- | -------------------------------------------------------------------------------------------------------------------------- |
+| `--job-id <id>`        | string | yes      | Import job ID returned by `cmodel import` (required)                                                                       |
+| `--follow`             | switch | no       | Block and poll until a terminal state. Without it, a single status probe is performed and the command returns immediately. |
+| `--interval <sec>`     | number | no       | Seconds between polls with --follow (default: 10, min: 1). Ignored without --follow.                                       |
+| `--poll-timeout <sec>` | number | no       | With --follow, stop polling after this many seconds (default: no limit). Ignored without --follow.                         |
+| `--api-key <key>`      | string | no       | API key                                                                                                                    |
+| `--base-url <url>`     | string | no       | API base URL                                                                                                               |
+
+#### Notes
+
+- This command always calls the public DashScope endpoint (https://dashscope.aliyuncs.com) in the Beijing region. The API is not deployed on workspace-specific base URLs (https://<workspace>.cn-beijing.maas.aliyuncs.com, HTTP 404), so the Profile base URL is not used for this request; workspace API keys are accepted.
+- Default (no --follow) is a NON-BLOCKING single status probe for agents/scripts. Terminal statuses are SUCCESSED and FAILED; an unknown server status is surfaced verbatim and treated as non-terminal.
+- A terminal FAILED status raises a normal CLI error (non-zero exit) with the server error_code preserved; SUCCEEDED and still-running statuses return 0. With --follow, exceeding --poll-timeout raises a timeout error.
+
+#### Examples
+
+```bash
+bl cmodel watch --job-id 937b-xxxx
+```
+
+```bash
+bl cmodel watch --job-id 937b-xxxx --output json
+```
+
+```bash
+bl cmodel watch --job-id 937b-xxxx --follow
+```
+
+```bash
+bl cmodel watch --job-id 937b-xxxx --follow --interval 5 --poll-timeout 3600
+```
diff --git a/skills/bailian-cli/reference/index.md b/skills/bailian-cli/reference/index.md
index b808afcc..1a215a16 100644
--- a/skills/bailian-cli/reference/index.md
+++ b/skills/bailian-cli/reference/index.md
@@ -38,6 +38,12 @@ Use this index for the skill-scoped quick index and global flags.
 | `bl batch download`             | API Key        | Download original Batch results/errors, optionally export CSV                                                      | [batch.md](batch.md)           |
 | `bl batch get`                  | API Key        | Get Batch task details, optionally following progress                                                              | [batch.md](batch.md)           |
 | `bl batch list`                 | API Key        | List one page of Batch tasks                                                                                       | [batch.md](batch.md)           |
+| `bl cmodel base-models`         | Console        | List base models that can be imported as custom models                                                             | [cmodel.md](cmodel.md)         |
+| `bl cmodel delete`              | API Key        | Delete a custom model and its Bailian-side import record                                                           | [cmodel.md](cmodel.md)         |
+| `bl cmodel get`                 | API Key        | Get details of a custom model (import job)                                                                         | [cmodel.md](cmodel.md)         |
+| `bl cmodel import`              | API Key        | Import a custom model from OSS (creates an import job)                                                             | [cmodel.md](cmodel.md)         |
+| `bl cmodel list`                | API Key        | List custom models (import jobs)                                                                                   | [cmodel.md](cmodel.md)         |
+| `bl cmodel watch`               | API Key        | Probe a custom model's import status (default: single non-blocking fetch). Pass --follow to poll until terminal.   | [cmodel.md](cmodel.md)         |
 | `bl config agent`               | No Auth        | Configure a coding agent to use DashScope API                                                                      | [config.md](config.md)         |
 | `bl config list`                | No Auth        | List config profiles and show the active profile                                                                   | [config.md](config.md)         |
 | `bl config set`                 | No Auth        | Set a config value                                                                                                 | [config.md](config.md)         |
@@ -126,6 +132,7 @@ Use this index for the skill-scoped quick index and global flags.
 | `app`        | `call`, `list`                                                                                                                                                                                                               | [app.md](app.md)               |
 | `auth`       | `generate-access-token`, `login`, `logout`, `status`                                                                                                                                                                         | [auth.md](auth.md)             |
 | `batch`      | `cancel`, `convert`, `create`, `download`, `get`, `list`                                                                                                                                                                     | [batch.md](batch.md)           |
+| `cmodel`     | `base-models`, `delete`, `get`, `import`, `list`, `watch`                                                                                                                                                                    | [cmodel.md](cmodel.md)         |
 | `config`     | `agent`, `list`, `set`, `show`, `ui`, `use`                                                                                                                                                                                  | [config.md](config.md)         |
 | `console`    | `call`                                                                                                                                                                                                                       | [console.md](console.md)       |
 | `decision`   | `(root)`                                                                                                                                                                                                                     | [decision.md](decision.md)     |
diff --git a/skills/bailian-finetune/SKILL.md b/skills/bailian-finetune/SKILL.md
index 455f3286..a958b2bc 100644
--- a/skills/bailian-finetune/SKILL.md
+++ b/skills/bailian-finetune/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: bailian-finetune
 metadata:
-  version: "2.7.2"
+  version: "2.7.3"
   requires:
     bins: ["bl"]
 description: >-
@@ -46,24 +46,25 @@ BAILIAN_CALLER_TYPE=agent BAILIAN_CALLER_NAME=codex BAILIAN_SKILL_NAME=bailian-f
 
 ## When to use which command
 
-| Intent                                 | Command                                                                                          |
-| -------------------------------------- | ------------------------------------------------------------------------------------------------ |
-| Validate / upload training data        | `bl dataset validate` / `upload` (`.jsonl` or `.zip`)                                            |
-| Dataset list / detail / delete         | `bl dataset list` / `get` / `delete`                                                             |
-| Create a fine-tuning job               | `bl finetune text\|audio\|image create`                                                          |
-| Job list / detail / follow             | `bl finetune list` / `get` / `watch` / `logs`                                                    |
-| Artifacts and export                   | `bl finetune checkpoints` / `export`                                                             |
-| Quantize / compress a full model       | `bl finetune compress templates` / `create` / `watch` / `get` / `logs` / `list`                  |
-| Cancel / delete a compression job      | `bl finetune compress cancel` / `delete`                                                         |
-| Cancel / delete a fine-tuning job      | `bl finetune cancel` / `delete`                                                                  |
-| Trainable capability lookup            | `bl finetune capability`                                                                         |
-| Deploy / lifecycle                     | `bl deploy text\|audio\|image create`, `list` / `get` / `update` / `scale` / `delete` / `models` |
-| Query throughput reservations          | `bl deploy list --plan ptu` / `bl deploy get`                                                    |
-| Query capacity instances               | `bl deploy capacity list` / `get`                                                                |
-| Query / wait for a capacity operation  | `bl deploy operation get` / `wait`                                                               |
-| Buy / scale / renew / release capacity | `bl deploy capacity create` / `scale` / `renew` / `delete`                                       |
-| Unsubscribe a prepaid instance         | `bl deploy capacity unsubscribe` (builds the billing console refund link)                        |
-| Configure ModelCode overflow strategy  | `bl deploy overflow`                                                                             |
+| Intent                                 | Command                                                                                                |
+| -------------------------------------- | ------------------------------------------------------------------------------------------------------ |
+| Validate / upload training data        | `bl dataset validate` / `upload` (`.jsonl` or `.zip`)                                                  |
+| Dataset list / detail / delete         | `bl dataset list` / `get` / `delete`                                                                   |
+| Create a fine-tuning job               | `bl finetune text\|audio\|image create`                                                                |
+| Job list / detail / follow             | `bl finetune list` / `get` / `watch` / `logs`                                                          |
+| Artifacts and export                   | `bl finetune checkpoints` / `export`                                                                   |
+| Quantize / compress a full model       | `bl finetune compress templates` / `create` / `watch` / `get` / `logs` / `list`                        |
+| Cancel / delete a compression job      | `bl finetune compress cancel` / `delete`                                                               |
+| Cancel / delete a fine-tuning job      | `bl finetune cancel` / `delete`                                                                        |
+| Import an external OSS LoRA model      | Hand-off: `bl cmodel import *` — hub-owned commands; see `bailian-cli` reference or `bl cmodel --help` |
+| Trainable capability lookup            | `bl finetune capability`                                                                               |
+| Deploy / lifecycle                     | `bl deploy text\|audio\|image create`, `list` / `get` / `update` / `scale` / `delete` / `models`       |
+| Query throughput reservations          | `bl deploy list --plan ptu` / `bl deploy get`                                                          |
+| Query capacity instances               | `bl deploy capacity list` / `get`                                                                      |
+| Query / wait for a capacity operation  | `bl deploy operation get` / `wait`                                                                     |
+| Buy / scale / renew / release capacity | `bl deploy capacity create` / `scale` / `renew` / `delete`                                             |
+| Unsubscribe a prepaid instance         | `bl deploy capacity unsubscribe` (builds the billing console refund link)                              |
+| Configure ModelCode overflow strategy  | `bl deploy overflow`                                                                                   |
 
 The `capacity list` / `get`, `operation get` / `wait` and `deploy list` / `get` queries are read-only. Capacity values are kTPM; effective, configured and target capacities are distinct. `deploy list --status` filters only the fetched page locally, with the server total left unfiltered. `operation get` reports status as data; `operation wait` exits non-zero on failure or timeout and refreshes capacity after success. Use IDs returned by the API; waiting never retries a write.
 
diff --git a/skills/bailian-gen/SKILL.md b/skills/bailian-gen/SKILL.md
index 039df61d..e69664f2 100644
--- a/skills/bailian-gen/SKILL.md
+++ b/skills/bailian-gen/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: bailian-gen
 metadata:
-  version: "2.7.2"
+  version: "2.7.3"
   requires:
     bins: ["bl"]
 description: >-
diff --git a/skills/bailian-knowledge/SKILL.md b/skills/bailian-knowledge/SKILL.md
index 06e7c667..91141e76 100644
--- a/skills/bailian-knowledge/SKILL.md
+++ b/skills/bailian-knowledge/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: bailian-knowledge
 metadata:
-  version: "2.7.2"
+  version: "2.7.3"
   requires:
     bins: ["bl"]
 description: >-
diff --git a/skills/bailian-managed-agent/SKILL.md b/skills/bailian-managed-agent/SKILL.md
index 1060e7a4..fffd13d6 100644
--- a/skills/bailian-managed-agent/SKILL.md
+++ b/skills/bailian-managed-agent/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: bailian-managed-agent
 metadata:
-  version: "2.7.2"
+  version: "2.7.3"
   requires:
     bins: ["bl"]
 description: >-
diff --git a/skills/bailian-memory/SKILL.md b/skills/bailian-memory/SKILL.md
index 4db43c77..b4832dad 100644
--- a/skills/bailian-memory/SKILL.md
+++ b/skills/bailian-memory/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: bailian-memory
 metadata:
-  version: "2.7.2"
+  version: "2.7.3"
   requires:
     bins: ["bl"]
 description: >-
diff --git a/skills/bailian-protocol/SKILL.md b/skills/bailian-protocol/SKILL.md
index 9e80dab3..31ce4cf0 100644
--- a/skills/bailian-protocol/SKILL.md
+++ b/skills/bailian-protocol/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: bailian-protocol
 metadata:
-  version: "2.7.2"
+  version: "2.7.3"
   requires:
     bins: ["bl"]
 description: >-
diff --git a/skills/bailian-sandbox/SKILL.md b/skills/bailian-sandbox/SKILL.md
index 7ac90955..6cd73985 100644
--- a/skills/bailian-sandbox/SKILL.md
+++ b/skills/bailian-sandbox/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: bailian-sandbox
 metadata:
-  version: "2.7.2"
+  version: "2.7.3"
   requires:
     bins: ["bl"]
 description: >-
diff --git a/skills/bailian-web-search/SKILL.md b/skills/bailian-web-search/SKILL.md
index e1869b90..d9d77410 100644
--- a/skills/bailian-web-search/SKILL.md
+++ b/skills/bailian-web-search/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: bailian-web-search
 metadata:
-  version: "2.7.2"
+  version: "2.7.3"
   requires:
     bins: ["bl"]
 description: >-