CSP课件制作技能

• Outcome: Pass
• Latest audit: May 29, 2026 at 8:41 PM UTC
• Version:
  1.0.0

Overview

This skill generates C++ teaching materials and its file, PDF, and dependency behavior mostly matches that purpose, but users should choose a clean output folder and fix the template paths before running it.

Install only in a dedicated workspace. Before running the Node scripts, replace the hardcoded C:/Users/ning/... output paths with your intended folder, review any Move-Item command so it only moves generated files, and remove the Google Fonts import if offline or privacy-preserving HTML output matters.

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
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration

Findings (6)

Description-Behavior Mismatch

Medium

• Confidence: 97% confidence
• Finding: The script writes output to a hardcoded absolute Windows path in a developer-specific directory. This can cause unauthorized or unexpected filesystem writes on the host running the skill, break portability, and overwrite files in an environment the user did not approve; in an agent skill context, implicit local writes are more concerning because execution may happen on shared or production hosts.

Context-Inappropriate Capability

Low

• Confidence: 94% confidence
• Finding: Hardcoded access to a personal working directory grants the skill unnecessary filesystem context beyond what is required to generate a presentation. Even without overtly malicious logic, unnecessary path assumptions increase the blast radius of execution errors and can expose host-specific structure or create files in sensitive locations.

Context-Inappropriate Capability

Low

• Confidence: 93% confidence
• Finding: The template loads Google Fonts from an external domain, which creates an unnecessary outbound network dependency for a locally generated teaching game. Even though this is not code execution, it can leak user metadata such as IP address, user agent, and access timing to a third party, and it also weakens offline/privacy-preserving use of the skill.

Vague Triggers

Medium

• Confidence: 77% confidence
• Finding: The trigger words include very broad everyday phrases like '做课件' and '做教案', which can cause accidental invocation in unrelated contexts. Mis-triggering is dangerous here because the skill then recommends dependency installation, file generation, directory manipulation, and PDF processing, expanding its operational footprint without clear user intent.

Vague Triggers

Medium

• Confidence: 80% confidence
• Finding: The scenario description claims applicability to nearly any C++ or competitive-programming teaching request, without firm boundaries or disallowed cases. In a skill that can read PDFs and write structured outputs, such ambiguity increases the chance of overbroad delegation, unintended processing of user files, and execution of risky workflow steps when a simpler answer would suffice.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The code performs a filesystem write to a fixed local path without any user-facing notice, consent, or runtime choice of destination. In an agent skill, silent local writes are risky because users may not expect host-side persistence, and the behavior can overwrite artifacts or leave files in locations outside the intended workspace.

VirusTotal

67/67 vendors flagged this skill as clean.