MLX Swift LM Expert

• Outcome: Pass
• Latest audit: May 28, 2026, 2:01 AM
• Version:
  1.0.0

Overview

This is a documentation-only Swift ML skill whose network, cache, credential, and tool-calling examples fit its stated purpose.

Install is reasonable if you want Swift MLX reference material. When using the examples, pin trusted model revisions where practical, keep Hugging Face tokens out of source code, restrict remote media URLs, and require validation or confirmation before connecting model tool calls to file changes, account actions, payments, or public posting.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (3)

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The documentation shows executing model-generated tool calls directly via `toolCall.execute(with: weatherTool)` without an explicit safety warning that tool calls are untrusted model output and may cause side effects, external requests, or data access. In LLM tooling workflows, readers often copy examples verbatim, so this omission can normalize unsafe execution patterns and increase the chance of prompt-injection-driven or unintended tool invocation in downstream applications.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The skill explicitly states that models are downloaded automatically from Hugging Face but does not warn users that invoking the example will trigger outbound network access and retrieval of third-party artifacts. In an agent context, undocumented network access can violate user expectations, leak metadata such as IP/project usage, and introduce supply-chain risk from unreviewed remote model content.

Missing User Warnings

Medium

• Confidence: 94% confidence
• Finding: The examples document `UserInput.Image.url(...)` and `UserInput.Video.url(...)` for file or remote URLs without cautioning that processing such inputs may cause the application to fetch attacker-controlled external resources. In agent or server environments this can enable SSRF-style behavior, unexpected egress, access to internal services, and privacy leaks through remote fetches.

VirusTotal

66/66 vendors flagged this skill as clean.