Security audit

Ai Website Manager

Security checks across malware telemetry and agentic risk

Overview

This is a coherent website-building guide, but it asks users to handle real deployment accounts and credentials, so it should be used carefully.

Install this only if you want an agent to help manage a real website. Prefer private repositories and preview deployments until ready, approve every commit/push/deploy/domain/CMS change, use expiring least-privilege tokens where possible, keep .env.local out of Git, and never paste passwords or secret tokens into chat.

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 README explicitly states that the skill triggers automatically for broad, ordinary website-related requests such as 'build a website' or 'my website is broken.' In an agent environment, this can cause the skill to activate in situations the user did not clearly consent to, expanding the skill's influence over account setup, deployment, and troubleshooting workflows in ways that may steer actions or data collection unexpectedly.

Vague Triggers

High

Confidence: 96% confidence
Finding: The manifest uses very broad automatic-trigger phrases such as 'help me build' and activation based on users seeming 'unsure/nervous,' which can cause the skill to activate during ordinary conversation unrelated to website work. Overbroad activation increases the chance of inappropriate context capture, undesired workflow steering, and privilege misuse by causing the agent to enter this skill when the user did not explicitly request it.

Natural-Language Policy Violations

Medium

Confidence: 91% confidence
Finding: The skill instructs the agent to speak Hebrew first if the user's name or context 'suggests they're Israeli,' which relies on nationality or ethnicity inference without consent. This creates privacy and profiling risks, can lead to discriminatory or incorrect behavior, and may expose sensitive attribute inference in situations where the user did not ask for language personalization.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The guide explicitly instructs users to create a Vercel token with no expiration and store it locally for automation, but it does not emphasize least privilege, token rotation, or the increased risk of long-lived deployment credentials. If the local environment, logs, backups, or developer machine are compromised, an attacker could use the token to manage deployments or project settings for an extended period.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The file tells users to create a Sanity API token with Editor permissions, which grants write access to CMS content, and store it in `.env.local` without highlighting the sensitivity of a write-capable credential. If exposed, an attacker could modify or delete site content, publish malicious content, or tamper with the CMS-backed website.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.