Security audit

Debug Root Cause

Security checks across malware telemetry and agentic risk

Overview

This looks like a debugging aid, but it should go through Review because it can activate broadly, write local notes, and points agents toward self-modifying workflows.

Install only if you are comfortable with a debugging skill that may activate on broad failure language, write local scratch notes, and direct the agent toward self-repair through another skill. Require explicit confirmation before any skill-manager, self-optimization, or skill-modification step, and avoid putting secrets or production data into debug notes.

SkillSpector

By NVIDIA

Vulnerability Patterns

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
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration

Findings (5)

Context-Inappropriate Capability

Medium

Confidence: 95% confidence
Finding: The skill explicitly instructs loading another skill ('skill-manager') to repair or optimize itself, which introduces self-modification and capability expansion beyond root-cause analysis. This creates an escalation path where routine debugging can trigger secondary skills and potentially broader behavior without clear user approval or safety boundaries.

Intent-Code Divergence

Medium

Confidence: 93% confidence
Finding: The document presents itself as a debugging methodology but later claims the skill can 'self-evolve' and should load another skill for repair or optimization. This contradiction hides behavior outside the declared scope, making it easier for users or agents to invoke modification workflows under the guise of normal debugging.

Vague Triggers

High

Confidence: 90% confidence
Finding: The trigger definition is overly broad and mandatory, causing the skill to activate on many common phrases or any unexpected tool result. Overbroad auto-invocation increases the chance that the agent performs unnecessary investigative actions, writes files, or follows this skill in situations where simpler or safer handling would be more appropriate.

Vague Triggers

Medium

Confidence: 84% confidence
Finding: The trigger table uses ambiguous phrases like 'why', 'debug', and 'still broken' without enough negative examples or disambiguation rules. This can cause accidental activation in normal conversation, leading to unnecessary procedural steps and expanding the skill's operational footprint.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: The skill instructs writing and later updating a file in /tmp as part of normal execution, but it does not warn that it is modifying the local filesystem or obtain approval. Even though /tmp is temporary, this still creates side effects, can overwrite existing files, and may expose debugging context or sensitive data in a shared environment.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.