huawei-cloud-modelarts-training-diagnosis

• Outcome: Review
• Latest audit: Aug 3, 2026 · 9:39 AM UTC
• Version: 1.0.0

Overview

The skill is a coherent read-only ModelArts diagnosis helper, but its documentation includes risky install, credential, log, and cleanup guidance that users should review before installing.

Install only if you are comfortable granting the agent read-only access to ModelArts training job metadata and diagnostic logs through hcloud. Use least-privilege Huawei Cloud credentials, do not paste or expose AK/SK values, redact logs and temporary OBS URLs before sharing, verify installer downloads independently instead of blindly running non-interactive scripts, and run rm -rf ~/.hcloud/ only if you intentionally want to remove local hcloud profiles and credentials after backing them up.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Tool Misuse: Tool Parameter Abuse, Chaining Abuse, Unsafe Defaults
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• YARA Signatures: Malware Match, Webshell Match, Cryptominer Match
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands

Findings (9)

Vague Triggers

Medium

• Confidence: 90% confidence
• Finding: The trigger set includes broad natural-language phrases such as generic requests to diagnose failed or stuck training jobs, which can cause the skill to activate in situations the user did not clearly intend. In this skill, unintended invocation is more sensitive because the workflow can progress into account-wide job enumeration and log retrieval, exposing operational metadata and potentially sensitive log snippets.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The workflow allows scanning all abnormal training jobs and retrieving previews or URLs for training logs, but it does not present a prominent upfront warning that these actions may access account-wide metadata and sensitive log content. Because training logs can contain stack traces, paths, host details, and secrets, users may unknowingly authorize broader data access than expected.

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: The acceptance criteria explicitly instruct running `hcloud configure list` and treating visible AK/SK as part of the success condition, but provide no warning about credential sensitivity, redaction, or safe handling. In a diagnostic skill context, encouraging display of configured access credentials can lead to accidental disclosure in terminal captures, logs, transcripts, screenshots, or agent output, which could enable unauthorized access to Huawei Cloud resources.

Missing User Warnings

Medium

• Confidence: 90% confidence
• Finding: The document directs users to retrieve training job log previews and OBS log URLs without any warning that logs and storage URLs may contain sensitive data such as stack traces, dataset paths, tokens, internal endpoints, or customer code details. In this skill's failure-diagnosis context, those artifacts are useful, but absent minimization or redaction guidance they create a real risk of over-collection and unintended disclosure through chat, persistent logs, or copied outputs.

Missing User Warnings

Medium

• Confidence: 97% confidence
• Finding: The guide instructs users to download and immediately execute a remote shell script, including a non-interactive mode that suppresses confirmation, without any integrity verification, script review, or warning. If the distribution endpoint, network path, or script content is compromised, this yields arbitrary code execution on the host with the user's privileges.

Missing User Warnings

Medium

• Confidence: 91% confidence
• Finding: The uninstall and cleanup section includes irreversible deletion commands that remove binaries, installation directories, config, and cache data without a clear warning about data loss or profile removal. In an agent skill context, such commands can be copied or automated blindly, causing unintended destruction of local credentials, history, or configuration state.

Missing User Warnings

Medium

• Confidence: 90% confidence
• Finding: The verification guide instructs users to check and initialize AK/SK-based authentication but does not include any warning about treating access keys as secrets, avoiding shell history exposure, or using least-privilege credentials. In a cloud operations skill, this omission can lead to unsafe handling of long-lived credentials, increasing the risk of credential leakage and unauthorized access to ModelArts and other cloud resources.

Tool Parameter Abuse

High

• Category: Tool Misuse

Content

# Linux/macOS manual uninstallation:
sudo rm -f /usr/local/bin/hcloud
sudo rm -rf /usr/local/hcloud/
rm -rf ~/.hcloud/

# Windows manual uninstallation:
# 1. Delete hcloud.exe file

• Confidence: 95% confidence
• Finding: rm -rf ~/.hcloud/

YARA rule 'agent_skill_destructive_autonomous_actions': Autonomous destructive filesystem, shell history, or repository actions in AI agent skills [agent_skills]

High

• Category: YARA Match

Content

# Linux/macOS manual uninstallation:
sudo rm -f /usr/local/bin/hcloud
sudo rm -rf /usr/local/hcloud/
rm -rf ~/.hcloud/

# Windows manual uninstallation:
# 1. Delete hcloud.exe file

• Confidence: 92% confidence
• Finding: rm -rf ~/; rm -rf ~/; Non-interactive

VirusTotal

VirusTotal findings are pending for this skill version.

Static analysis

Detected: suspicious.destructive_delete_command

Documentation contains a destructive delete command without an explicit confirmation gate.

Warn

• Code: suspicious.destructive_delete_command
• Location: references/cli-installation-guide.md:212