Security audit

PPT 演示工具包基础版

Security checks across malware telemetry and agentic risk

Overview

This PowerPoint skill is mostly a document-editing helper, but its privacy and network behavior are inconsistent enough that users should review it before installing.

Install only if you are comfortable reviewing each operation before it runs. Use copies of presentations, avoid callback_url for confidential files, and do not rely on the local-only privacy claim unless the executing agent and tools are configured to block network access.

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 (7)

Description-Behavior Mismatch

Medium

Confidence: 94% confidence
Finding: The skill presents itself as local PPTX processing, but the documented `callback_url` and dependency on external APIs/network access create a mismatch in data-flow expectations. Users may supply sensitive presentation content assuming it remains local, when job metadata or content could be transmitted off-host during execution or callbacks.

Description-Behavior Mismatch

High

Confidence: 98% confidence
Finding: The documentation states that all FREE-version data is stored locally and not uploaded, while other sections say some functionality requires network access or external APIs. This contradiction can mislead users into exposing confidential presentation data under false privacy assumptions, which is especially risky for business documents.

Intent-Code Divergence

High

Confidence: 97% confidence
Finding: Conflicting statements about whether data is purely local or may traverse the network undermine informed consent and safe operation. In a document-processing skill, this is dangerous because presentations often contain proprietary or personal information, and operators may make trust decisions based on the privacy claims in the skill.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The trigger conditions mention broad data-analysis and reporting scenarios that do not match the stated PPTX-specific scope. Overbroad triggers increase the chance the agent invokes this skill for unrelated tasks, causing inappropriate file handling, unnecessary command execution, or accidental disclosure through the wrong workflow.

Vague Triggers

Medium

Confidence: 84% confidence
Finding: The natural-language invocation guidance is generic enough that an agent may route loosely related requests into this skill without strong constraints. In an `exec`-capable skill, ambiguous routing can expand the attack surface by encouraging execution in contexts the user did not intend.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The skill supports editing and creation of presentation files but does not clearly warn that operations may modify or overwrite user content. That omission increases the risk of destructive changes, especially when combined with agent automation and broad natural-language triggering.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: Accepting a `callback_url` without a nearby warning about outbound transmission creates a hidden exfiltration channel for job status, metadata, or potentially user-provided content. In a presentation-processing context, even metadata such as filenames, titles, and processing results may be sensitive.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.