Security audit

密钥卫士

Security checks across malware telemetry and agentic risk

Overview

This key-protection skill needs Review because it asks to handle API keys with broad file, command, write, API-call, and logging behavior that is not tightly scoped.

Review carefully before installing. Only use this skill where you are comfortable letting an agent inspect API-key-related content, environment/configuration data, and scripts, and require explicit confirmation before any file write, command execution, API call, or logging of security events.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
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
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (6)

Intent-Code Divergence

High

Confidence: 96% confidence
Finding: The examples instruct the agent to validate keys, call external APIs, read masked files, and write local files, which materially expands the skill from a passive key-guardrail into an active integration and file-modification workflow. This mismatch can cause the agent to perform network and filesystem actions under a vague security-themed label, increasing the chance of unintended secret handling, unauthorized outbound requests, or unsafe file edits.

Intent-Code Divergence

Medium

Confidence: 88% confidence
Finding: The FAQ claims capabilities like origin checking, interception logging, and integration with other security tools, but the skill is described as markdown-driven and does not define a trustworthy mechanism for enforcing those controls. Overstated security functionality can create false assurance, leading users to rely on protections that do not actually exist while exposing secrets or requests they believed were being guarded.

Vague Triggers

High

Confidence: 91% confidence
Finding: A trigger like 'when user asks to call' is so broad that it can activate on many ordinary requests, potentially causing the skill to intercept, transform, or route actions in contexts the user did not intend. In a skill with read/exec/write tools and secret-related behavior, ambiguous invocation criteria increase the risk of overreach and accidental sensitive operations.

Vague Triggers

High

Confidence: 93% confidence
Finding: The repeated invocation rule 'when user asks to call' lacks specificity about what is being called, what data is in scope, and whether secrets or local files may be touched. This ambiguity is especially dangerous here because the skill presents itself as a guardrail while also advertising operational capabilities that can change files and reach external services.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The skill description does not clearly warn users that the skill may read files, execute commands, or make external API-related actions despite declaring those tools and workflows elsewhere. Missing disclosure undermines informed consent and can cause users to expose local data or authorize risky operations under the mistaken belief that the skill only performs passive safety checks.

Missing User Warnings

Medium

Confidence: 84% confidence
Finding: The examples mention interception logs for API keys without explaining minimization, retention, masking, access controls, or whether full secrets are ever stored. Logging secret material, even for defensive reasons, can itself become a leakage vector if logs are persisted, shared, or insufficiently redacted.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.