安全加固之盾

• Outcome: Review
• Latest audit: Aug 20, 2026 · 4:57 AM UTC
• Version: 1.0.1

Overview

This security-audit skill is mostly purpose-aligned, but it leaves command execution, report writing, external callbacks, and API-key expectations too broad or inconsistent for automatic trust.

Install only if you intend to let an agent inspect a codebase and optionally run security scanners. Before use, require explicit confirmation for each command, choose where reports are written, avoid sending findings to callback URLs unless you trust the destination, and clarify which third-party scanners or API keys will be used.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Behavioral AST: exec() Call, eval() Call, Dynamic Import
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands

Findings (6)

Intent-Code Divergence

Medium

• Confidence: 96% confidence
• Finding: The documentation gives conflicting guidance about whether API keys are required, which can cause users to misconfigure secrets handling or assume a feature works without proper authentication. In a security-focused skill, contradictory secret-management instructions are especially risky because they can lead to insecure setup, accidental credential exposure, or unsafe operational assumptions.

Description-Behavior Mismatch

Medium

• Confidence: 91% confidence
• Finding: The manifest presents the skill as narrowly scoped to security hardening, but the body advertises much broader automation behavior. This scope mismatch can mislead users and downstream agents into granting broader trust or execution latitude than intended, increasing the chance of unsafe tool use under the guise of a focused security skill.

Context-Inappropriate Capability

Medium

• Confidence: 94% confidence
• Finding: The skill claims generic file handling, external API integration, command execution, and information retrieval capabilities that are not justified by its stated security-hardening purpose. In the context of a skill with read and exec tools, this broadening materially increases attack surface by normalizing unrelated operations that could exfiltrate data, run unsafe commands, or manipulate files beyond user expectations.

Vague Triggers

Medium

• Confidence: 88% confidence
• Finding: The invocation description is broad and ambiguous, making it unclear when the skill should be used and what actions it may take. Ambiguity is dangerous for agent skills because it can cause over-invocation in inappropriate contexts, including sensitive repositories or workflows where exec and report generation were not intended.

Missing User Warnings

Low

• Confidence: 84% confidence
• Finding: The skill describes generating multiple report files but does not clearly warn users that files will be written. Hidden write behavior can surprise users, alter repositories, leak findings into tracked files, or create sensitive artifacts in insecure locations, though the impact here is lower than direct code execution.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The documentation normalizes use of exec and third-party scanning tools without a clear, prominent warning about command execution, prerequisites, side effects, or trust boundaries. In a skill that can operate on codebases and secrets, this is dangerous because users may unknowingly permit commands that inspect sensitive data, invoke networked tools, or modify environments.

VirusTotal

VirusTotal findings are pending for this skill version.

Static analysis

No suspicious patterns detected.