Huawei Cloud ModelArts Skill

Manage Huawei Cloud ModelArts resources including training jobs, models, inference services, notebooks, and resource overview with secure temporary credentials.

Security Scan

The skill appears to implement ModelArts resource management as claimed, but its runtime behavior is vague about where credentials come from and it can read/upload local code directories (possible exfiltration) without documenting parameters or required environment context.

Purpose & Capability

Name/description, requirements.txt (modelarts-sdk), and implemented functions (list training jobs, create training job, list models/services/notebooks) are consistent with a ModelArts management skill. Declared SKILL_METADATA states support_auth_type 'system_token', which aligns with cloud SDK usage.

Instruction Scope

SKILL.md describes supported actions but does not document parameters or explain that create_training_job will take a local source_directory (code_dir) and boot_file. scripts/skill.py accepts a source_directory param and passes it to Estimator(...) which typically packages and uploads local files — meaning the agent may read arbitrary local directories and upload them to cloud. The documentation claims 'no persistence' and 'in-memory only' but offers no explicit guarantees or constraints about what is read/uploaded. The skill's runtime instructions are therefore vague and grant the agent discretion to access local filesystem content.

Install Mechanism

No install spec is provided (instruction-only installation), but the package includes code files and requirements.txt pointing to modelarts-sdk from PyPI. This is an expected dependency for ModelArts integration and is not a suspicious download pattern; however the presence of code (scripts/skill.py) means the skill will execute bundled Python, so review of code is required before trust.

Credentials

The skill lists no required environment variables yet calls modelarts.session.Session() to obtain runtime credentials implicitly (SKILL_METADATA mentions 'system_token' auth). That is plausible for code running inside Huawei's ModelArts environment, but the lack of explicit declaration means the skill will try to access whatever instance/host credentials are available (metadata/system token). Also, create_training_job accepts a local source_directory parameter which may cause sensitive local files to be uploaded — the skill requests access (implicitly) to filesystem and cloud credentials without documenting constraints.

Persistence & Privilege

always is false, the skill does not attempt to modify other skills or system-wide config, and its scripts claim no caching/persistence. There is no install mechanism that writes into non-standard system paths. However, the skill will interact with cloud resources (OBS, training services) and may transfer local files to cloud as part of job creation.

What to consider before installing

This skill mostly does what it says (ModelArts management), but before installing or invoking it consider: 1) Only run it in an appropriate ModelArts/cloud environment where the implicit system token behavior is intended — it will attempt to obtain runtime credentials via the SDK. 2) Be careful with create_training_job: it accepts a source_directory (code_dir) parameter that will likely read and upload local files; do not point it at directories containing secrets or sensitive files. 3) Ask the author to document required parameters (code_dir/boot_file/job_name), exactly how credentials are obtained, and to add explicit safeguards (whitelisted paths, explicit confirmation before uploading). 4) If you must use it outside a controlled cloud environment, run it in an isolated sandbox and review the package contents and the modelarts-sdk behavior for how credentials are sourced and temporary artifacts are handled. 5) The skill's desensitization is basic (name-based masking) and may not reliably remove all secrets from returned data — do not rely on it as the sole protection.

# ModelArts 资源全功能管理 Skill

## 基本信息

- 技能名称：ModelArts_Resource_Manager

- 版本：1.0.0

- 认证方式：ModelArts 运行环境临时安全凭证

- 安全等级：高

## 功能说明

提供 ModelArts 全栈基础资源管理能力：

- 资源概览查询

- 训练作业管理

- 模型管理

- 推理服务管理

- Notebook 管理

## 安全特性

- 自动脱敏：AK/SK、网络信息、密钥全部屏蔽

- 无缓存、无存储、无持久化

- 异常不暴露内部结构

- 全程内存运行

## 支持 Action

| action | 说明 |

|--------|------|

| list_resource_overview | 查询资源概览 |

| list_training_jobs | 查询训练作业 |

| create_training_job | 创建训练作业 |

| list_models | 查询模型 |

| list_services | 查询推理服务 |

| list_notebooks | 查询 Notebook |

## 调用示例

{

  "action": "list\_resource\_overview",

  "params": {}

}