Security audit

Hizal Write

Security checks across malware telemetry and agentic risk

Overview

This skill has a clear memory-writing purpose, but it tells the agent to save information automatically and broadly without enough user control or sensitive-data safeguards.

Install only if you intentionally want an agent to create durable memory while it works. Before using it, prefer explicit save approval, narrow default scope, and rules forbidding storage of secrets, credentials, personal data, customer content, confidential code, or temporary reasoning notes unless deliberately approved.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (5)

Vague Triggers

High

Confidence: 94% confidence
Finding: The skill is explicitly self-triggering on very common conversational phrases and broad moments of learning, which makes activation likely during ordinary agent dialogue rather than only when persistence is clearly warranted. In a memory-writing skill, that overbroad trigger surface materially increases the chance of unsolicited persistence of sensitive context, user data, or transient reasoning-derived observations.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: The description directs continuous persistence of learned information but provides no privacy, retention, or data-minimization safeguards. Because the skill is designed to write memory and knowledge artifacts, omission of such guardrails predictably leads to retention of information that should remain ephemeral, including user-provided sensitive content or internal-only context.

Ssd 3

Medium

Confidence: 95% confidence
Finding: A self-triggering instruction to persist what the agent 'learns' continuously creates a broad path for capturing user-provided and context-derived sensitive information. Since the skill does not distinguish safe durable knowledge from sensitive session data, exploitation or normal use could result in persistent storage of information that should not be retained.

Ssd 3

Medium

Confidence: 93% confidence
Finding: The imperative 'Write as you build. Not optional.' encourages indiscriminate logging behavior and removes operator discretion to avoid storing sensitive or irrelevant context. In this skill's context, mandatory persistence increases the chance that confidential observations, internal context, or accidental data exposure will be written into long-lived memory stores.

Ssd 3

Medium

Confidence: 91% confidence
Finding: The skill directs use of 'full content' plus source file and source lines without any masking or sensitivity controls, which can preserve exact sensitive material and make later retrieval easier. That design increases both the amount of sensitive data retained and the fidelity of any leaked or misused memory entry.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.