modelarts-skill

ModelArts Skill

Huawei Cloud ModelArts platform integration skill, featuring modular design, OBS as Notebook primary storage, and environment variable auto-authentication.

---

🔐 Security Model (CRITICAL)

All credentials handled internally, never returned to LLM:

┌─────────────────────────────────────────────────────────────┐
│                     LLM / OpenClaw                          │
│  ❌  NEVER sees: AK, SK, Security Token, raw credentials     │
│  ✅  ONLY sees: Masked status, API results, resource IDs     │
└─────────────────────────────────────────────────────────────┘
                            ↕ (Safe data only)
┌─────────────────────────────────────────────────────────────┐
│              auth_manager.py (Secure Vault)                 │
│  🔒  Credentials stored in memory only                       │
│  ✅  Only masked data exposed externally                      │
└─────────────────────────────────────────────────────────────┘
                            ↕ (Secure session)
┌─────────────────────────────────────────────────────────────┐
│              Module Functions (Secure SDK Wrapper)          │
│  Unified pattern: access.sdk().execute(api_func, ...)        │
│  Returns: Standard format API responses only                 │
└─────────────────────────────────────────────────────────────┘

---

🔑 Authentication Methods

---

📚 Module Routing

Use keywords to find the corresponding reference file for function index and locations.

| Keyword | Reference File | |---------|----------------|---------------|--------| | notebook, jupyter, image, obs, flavor, cluster, auth | references/notebook.md | | pool, resource pool, dedicated pool, node, workload, os network, plugin | references/pool.md | | node pool, nodepool, scale node, node group, plugin, config template | references/node_pool.md | | train, training job, 训练任务 | references/train.md | | infer1.0, inference, service, 旧版推理服务,旧版在线服务 | references/infer_v1.md | | infer2.0, inference, service, 新版推理服务，在线服务,新版在线服务 | references/infer_v2.md | | management, workspace, authmode, authorization, quota, tag, scheduled event | references/management.md | | swr, image repo, 镜像仓库 | references/swr.md | | vpc, virtual private, subnet, security group, 虚拟私有云 | references/vpc.md | | kms, key management, 密钥管理 | references/kms.md | | liteserver, dev server, hyper cluster, hyperinstance, eip, roce | references/liteserver.md | | common, auth, api helper, format result, decorator | references/common.md |

---

📦 Module Architecture

modelarts/
├── scripts/
│   ├── common_module/          # 跨模块共享基础设施
│   │   ├── __init__.py
│   │   ├── _bootstrap.py       # 统一引导
│   │   ├── auth.py             # 认证管理
│   │   ├── api_helper.py       # API 调用封装
│   │   └── result.py           # 返回格式统一
│   │
│   ├── notebook_module/        # Notebook & 镜像管理
│   ├── pool_module/            # 资源池管理
│   ├── node_pool_module/       # 节点池管理
│   ├── train_module/           # 训练任务管理
│   ├── infer_v1_module/        # 旧版推理服务管理
│   ├── infer_v2_module/        # 新版推理服务管理
│   ├── management_module/      # 平台管理: 工作空间/授权/配额/事件
│   ├── swr_module/             # SWR 镜像仓库
│   ├── vpc_module/             # VPC 虚拟私有云
│   ├── kms_module/             # KMS 密钥管理服务
│   ├── liteserver_module/      # Lite Server 轻量计算节点管理 (含 HPS/超节点完整支持)
│   └── __pycache__/
│
└── references/
    ├── notebook.md
    ├── pool.md
    ├── node_pool.md
    ├── train.md
    ├── infer_v1.md
    ├── infer_v2.md
    ├── management.md
    ├── swr.md
    ├── vpc.md
    ├── kms.md
    └── common.md

---

✅ Design Features

1. Modular Design - One file per function, clean structure, easy to maintain
2. OBS as Primary Storage - Notebook can use OBS buckets directly as root filesystem
3. OBS Bucket Management - Built-in create/delete/list OBS bucket functions
4. Environment Variable Auth - Zero configuration for local/CI environments
5. Unified API Pattern - All functions use consistent calling pattern
6. Clean Deliverables - No temp scripts, no debug code, no redundant files
7. Node Pool Management - Independent node pool lifecycle management
8. Plugin Management - Resource pool plugin listing and creation (ListPoolPlugins, CreatePoolPlugin)
9. Node Configuration Template Query - Query node configuration template details (ShowNodeConfigTemplate)
10. Common Infrastructure - Extracted shared infrastructure, improved code reuse
11. Training Job Management - Full lifecycle: create/list/stop/delete training jobs
12. Inference Service Management - Old v1 API: create/list/start/stop/update/delete inference services
13. Workspace & Authorization - Workspace CRUD, auth mode, IAM authorization, quotas, scheduled events
14. SWR Image Repository - Query training images from SWR
15. VPC Virtual Private Cloud - Query VPCs, Subnets, Security Groups from Huawei Cloud VPC service
16. KMS Key Management Service - Query SSH keypairs from Huawei Cloud KMS
17. Lite Server Management - Full lifecycle management of lightweight compute nodes (51 APIs covering Chapter 22)
18. Hyper Cluster Management - Create, list, delete hyper clusters for high-performance computing
19. Hyperinstance Operations - Scale, start, stop, and manage hyper instances
20. Batch Operations - Bulk actions on multiple Lite Server instances (start/stop/reboot/changeOS/reinstallOS/delete)
21. Job Management - Create, list, delete jobs on Lite Server instances

---

🎯 Quick Reference by Use Case