Security audit

dgngjx-skill

Security checks across malware telemetry and agentic risk

Overview

This toolbox is mostly purpose-aligned, but it needs review because it can persist user content, process local files broadly, and upload meeting audio without strong privacy guardrails.

Install only if you are comfortable with a broad local toolbox. Avoid unattended --yes mode for sensitive or Office files, review outputs before sharing reports, clear or disable history for confidential work, and use local/manual meeting modes unless you explicitly trust the transcription API endpoint.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (6)

Intent-Code Divergence

Medium

Confidence: 95% confidence
Finding: The documented policy says Office-document access should require a warning and user confirmation, but the provided `_is_safe_file()` logic treats `assume_yes=True` as fully safe for Office files. Because `DGNGJX_ASSUME_YES=1` or `--yes` can be set in unattended contexts, the skill can silently read/write potentially sensitive Office documents despite the stated safety boundary.

Intent-Code Divergence

High

Confidence: 98% confidence
Finding: The skill claims a central safety filter that blocks dangerous file types and directories, but many file-processing snippets open paths directly without calling `_is_safe_file()` or `_is_safe_output()`. This creates a gap between the advertised protection and real behavior, allowing unsupported or sensitive files to be processed if a tool is invoked directly or routed incorrectly.

Missing User Warnings

High

Confidence: 96% confidence
Finding: The history feature stores user inputs and outputs persistently on disk under the user's home directory, yet the documentation does not present a clear privacy warning or consent flow. Because inputs/outputs may include sensitive text, file paths, business data, or secrets, this creates a real data-retention and local exposure risk.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The report generator consumes stored history entries and may re-display or summarize previously captured inputs and outputs without warning that historic sensitive content could resurface. This increases the chance of accidental disclosure in generated reports or shared documents.

Missing User Warnings

High

Confidence: 98% confidence
Finding: The online meeting-transcription mode uploads raw audio to an external API, but the documentation does not provide a clear privacy warning about off-device transmission of potentially confidential conversations. Meeting audio often contains sensitive business, personal, or regulated information, so silent external upload materially raises exposure risk.

Ssd 3

Medium

Confidence: 94% confidence
Finding: The history and report features are designed to persist and later aggregate user inputs/outputs, which semantically encourages accumulation of potentially sensitive data. In a toolbox skill that may process files, hashes, transcripts, URLs, and user text, such retention expands the blast radius of any local compromise or accidental sharing.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.