ClawHub
Back to skill

Security audit

ISAI Mermaid Diagrams

Security checks across malware telemetry and agentic risk

Overview

The skill is a real Mermaid diagram helper, but it sends diagram contents to an external renderer by default without a clear user-facing privacy warning.

Use this only for non-sensitive diagrams unless you are comfortable sending the diagram source to mermaid.ink. For internal architecture, network, authentication, API, or data-flow diagrams, prefer a local Mermaid renderer or redact hostnames, service names, secrets, and internal topology before rendering.

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 (4)

Description-Behavior Mismatch

Medium
Confidence
97% confidence
Finding
The skill claims to generate Mermaid PNGs but actually instructs uploading the full diagram source to the external mermaid.ink service. Diagram content can contain sensitive architecture, network, auth, or data-flow details, so this creates an undocumented data exfiltration path and misrepresents how processing occurs.

Context-Inappropriate Capability

Medium
Confidence
92% confidence
Finding
A diagram-generation skill does not inherently require third-party network transmission, yet the workflow depends on an external web renderer. In this context, users may ask for internal system, cloud, security, or API-flow diagrams, making the transmitted content potentially sensitive and increasing the security risk.

Vague Triggers

Medium
Confidence
84% confidence
Finding
The trigger phrases are broad enough to activate on many generic requests involving drawing or diagrams. Because this skill performs file writes and network-based rendering, unintended invocation could cause sensitive user content to be processed by the skill and potentially transmitted externally without the user specifically requesting that behavior.

Missing User Warnings

Medium
Confidence
98% confidence
Finding
The skill explicitly sends diagram contents to mermaid.ink via `curl` but does not instruct the agent to warn the user that their diagram source leaves the local environment. For architecture and sequence diagrams, the source may reveal internal hosts, services, trust boundaries, credentials flow, or other confidential design details.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.