Security audit

comm-skill-tool-free

Security checks across malware telemetry and agentic risk

Overview

This communication-drafting skill is mostly coherent, but it handles sensitive conversation content while giving conflicting privacy and external-processing disclosures.

Review this skill carefully before installing. Avoid providing secrets, regulated data, or highly sensitive private conversations unless you are comfortable with possible LLM/API or callback processing. Do not use the broad environment-variable inspection example, and treat generated messages as drafts that you review before sending.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
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

Findings (6)

Intent-Code Divergence

Medium

Confidence: 91% confidence
Finding: The documentation says the tool will not send messages, but the skill also supports callback URLs and external API/tool invocation. That mismatch can mislead users into providing sensitive conversation content under a false assumption that nothing leaves the local environment, creating a data disclosure risk.

Intent-Code Divergence

High

Confidence: 97% confidence
Finding: The skill claims all data is stored locally and not uploaded to the cloud, yet elsewhere states that some features require network access and LLM/API support. For a communication tool handling conversation history and emotional signals, this contradiction materially increases the chance of sensitive user data being transmitted externally without informed consent.

Context-Inappropriate Capability

Low

Confidence: 88% confidence
Finding: The skill instructs users to enumerate environment variables matching API, KEY, TOKEN, or SECRET, which is unnecessary for drafting communication messages. Even with masking, this encourages secret discovery behavior and may normalize exposing or handling sensitive credentials in an unrelated workflow.

Vague Triggers

Medium

Confidence: 79% confidence
Finding: The activation model is described as generic natural-language requests without clear trigger boundaries, making it easier for the skill to engage on ambiguous prompts. In an agent environment with exec and network-capable tooling, vague invocation semantics can lead to unintended processing or tool use on sensitive inputs.

Vague Triggers

Medium

Confidence: 80% confidence
Finding: Telling users to simply describe their needs in conversation is too broad for a skill with flexible agent-driven execution. This increases the risk of accidental invocation, over-collection of context, or unintended use of tools when users are not clearly opting into the skill's behavior.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The description says the skill processes dialogue history and emotional cues but does not provide a clear privacy warning, despite later acknowledging network/API usage. Because this content is inherently sensitive, the omission can cause users to share personal or confidential communications without understanding storage, transmission, or retention implications.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.