Security audit

huawei-cloud-msmodelslim-model-analysis

Security checks across malware telemetry and agentic risk

Overview

The skill is mostly a coherent msModelSlim model-analysis guide, but it includes an under-warned troubleshooting path that can execute model-provided Python code.

Install only if you intend to use it for msModelSlim-focused model analysis. Before following troubleshooting steps, avoid `trust_remote_code=True` unless the model repository is trusted and reviewed, and prefer static inspection of `config.json` and model code when possible. Ask the agent to confirm before package upgrades or downloading model files.

SkillSpector

By NVIDIA

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: The manifest description says the skill determines whether a model is implemented via transformers or model-local code, and the workflow requires checking local modeling files as a supported analysis path. However, the Scope section explicitly marks 'Non-transformers implementations' as 'Not supported,' which conflicts with the documented behavior of analyzing model-local implementations to assess feasibility.

Description-Behavior Mismatch

Medium

Confidence: 92% confidence
Finding: The manifest describes a focused pre-adaptation analysis skill: identify implementation source, structural features, layer-by-layer loading requirements, and MoE fused weight risks. The later 'Enhanced Features' section expands behavior to migration blocker detection, operator support analysis, Ascend NPU coverage reporting, warnings, and compatibility scoring, which are broader platform compatibility and migration assessment capabilities not stated in the manifest description.

Vague Triggers

Medium

Confidence: 92% confidence
Finding: The manifest trigger list includes broad phrases such as "adapter", "feasibility", "model assessment", and generic model-analysis terms without clear constraints or exclusion conditions. In a markdown/manifest context, this can cause unintended invocation during ordinary conversations about models rather than specifically requesting this skill.

Vague Triggers

Medium

Confidence: 84% confidence
Finding: Examples like "Check if this model can be quantized" and "Assess model adaptation feasibility" are not tightly scoped to msModelSlim or this workflow, and the document does not pair them with exclusions. That ambiguity increases the chance the skill is invoked for unrelated quantization or model-review tasks.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: This markdown file includes a code example that instructs users to load a model with `trust_remote_code=True`, which can execute repository-provided Python code. The surrounding text does not disclose the safety implications or advise users to trust the model source before using this option.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.