Security audit

gbrain-guide

Security checks across malware telemetry and agentic risk

Overview

This skill is a coherent GBrain guide, but it asks agents to move and import local knowledge-base files and run admin repair actions without clear consent or rollback safeguards.

Install only if you intend agents to manage a GBrain knowledge base. Before letting an agent use it, require it to show the exact source paths, destination prefixes, files to import, MCP write tools, and any admin maintenance command it plans to run; prefer copy or dry-run workflows and keep backups for large imports.

SkillSpector

By NVIDIA

Vulnerability Patterns

Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration

Findings (5)

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The README instructs an agent to move local files into specific path prefixes and import them into GBrain, but it does not require explicit user confirmation, backups, dry-run behavior, or scope restrictions before modifying the filesystem. In an agent-execution context, these instructions could cause unintended reorganization of a user's knowledge base, overwriting conventions, or broad ingestion of sensitive local data beyond what the user expected.

Missing User Warnings

Medium

Confidence: 89% confidence
Finding: The document recommends running an automatic repair/onboarding operation with admin privileges as part of routine health governance, but it does not warn that such a command may change schema mappings, links, metadata, or other repository state. In a skill meant to guide autonomous tool use, this can normalize impactful administrative actions without ensuring the user understands or approves the resulting modifications.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The activation description contains broad triggers such as '把资料存进 GBrain / 用 GBrain 检索 / 整理知识库 / gbrain MCP 调用', which can match many routine knowledge-management requests and cause this skill to load in situations where the user did not clearly ask for GBrain operations. Because the skill includes write/import and file-reorganization guidance, overbroad activation increases the chance of unnecessary exposure to state-changing instructions and unintended local data modification.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: This section instructs the agent to import content into a local knowledge base using write-capable tools like sources_add, entry_import_content, and put_page, but it does not require explicit user confirmation before modifying persistent local data. In practice, that can lead an agent to perform irreversible or hard-to-audit ingestion actions on the user's machine based on an ambiguous request.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The workflow explicitly recommends moving files into classification-prefix directories, which alters the user's local filesystem layout, but it does not warn that this changes file organization or require consent and safeguards. That is risky because file moves can disrupt other tools, break references, or relocate sensitive material into indexed locations without the user realizing the consequences.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.