About this skill

Problem

ModelArts operations often scatter across notebook, pool, node pool, training, inference, SWR, and workspace APIs. Local CI and cloud notebooks also use different credential patterns, causing AK/SK, PROJECT_ID, and REGION to leak into ad-hoc scripts.

How it works and limits

This skill organizes those operations into routed modules: notebook, pool, node_pool, train, infer_v1, infer_v2, management, swr, and common. Notebook Mode detects the ModelArts environment automatically; Local Mode reads MODELARTS_AK, MODELARTS_SK, MODELARTS_PROJECT_ID, and MODELARTS_REGION. Credentials stay internal and are not returned to the LLM.

Coverage:
- Notebook: create, start, stop, delete, save images, attach/detach OBS storage, and manage OBS buckets
- Pools: dedicated pools, batch node operations, node pool scaling, plugin and node config template queries
- Training/inference: training job lifecycle, v1 and v2 inference services, versions, events, and health monitoring
- Management/SWR: workspaces, authorization, quotas, scheduled events, and training image queries

It is an ops API integration, not a replacement for console diagnostics. Verify region, quota, and IAM permissions before running mutating operations.

Screenshots

Use Cases

Use ModelArts credentials from environment variables in local or CI jobs to create and track training tasks.
Manage notebooks by creating, starting, stopping, deleting instances, and attaching OBS buckets as storage.
Operate dedicated pools with batch node reboot, lock, resize, and node configuration template queries.
Maintain inference services by listing v1/v2 services, starting/stopping/updating them, and checking events.