Security audit

unified-review

Security checks across malware telemetry and agentic risk

Overview

This review-orchestration skill is mostly coherent, but it can activate on very broad words and automatically run a project-local Python script without asking first.

Install only if you want broad review requests to invoke this orchestration flow. Before use, consider removing the generic triggers and changing deferred_capture execution to a confirm-before-run or dry-run step.

SkillSpector

By NVIDIA

Vulnerability Patterns

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
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Context-Inappropriate Capability

Medium

Confidence: 96% confidence
Finding: The skill instructs the agent to automatically execute a local Python script to persist backlog findings without prompting the user. That creates an implicit side effect on the local repository and expands a review-only workflow into arbitrary project-state modification, which is risky because review findings can be influenced by untrusted code/content under analysis.

Context-Inappropriate Capability

Low

Confidence: 80% confidence
Finding: The skill directs the agent to track progress via TodoWrite and references evidence files being written, introducing stateful writes beyond pure analysis. While less severe than arbitrary command execution, it still normalizes modifying local state during a review and may lead to unintended file/task artifacts without clear consent boundaries.

Vague Triggers

Medium

Confidence: 90% confidence
Finding: The trigger phrases are broad and generic, including terms like 'review' and 'analysis', which are likely to match ordinary user requests. Overbroad activation increases the chance that this orchestration skill runs unexpectedly and brings along its side effects, including agent dispatch and local write behaviors.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The skill documents automatic backlog-writing behavior but does not warn the user that project data may be modified. Silent modification is dangerous in a review context because users reasonably expect analysis, not automatic persistence of derived findings into local files or systems.

Shadow Command Trigger

Medium

Category: Trigger Abuse
Confidence: 88% confidence
Finding: Using 'review' as a trigger conflicts with a built-in command and can shadow or intercept expected platform behavior. In context, this is more dangerous because the skill is not purely informational: accidental activation could cause multi-agent execution and downstream state-changing actions.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.