Code Review Expert

• Outcome: Review
• Latest audit: May 27, 2026, 10:19 PM
• Version:
  1.0.0

Overview

This appears to be a code-review skill, but it may send full code and review outputs through multiple LLM steps without clear data-handling notice.

Install only if you are comfortable with submitted code and review artifacts being processed by the configured LLM provider. Avoid using it on repositories containing secrets, regulated data, or proprietary code unless you have confirmed provider, retention, redaction, and local-model options.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep

Findings (4)

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The review method sends user-supplied code into an LLM-driven multi-agent workflow without any explicit user-facing notice, consent, or data-handling boundary. In a code review skill, submitted code may contain proprietary logic, secrets, credentials, or customer data, so undisclosed transmission to an external model provider creates a real confidentiality and compliance risk.

Missing User Warnings

Medium

• Confidence: 98% confidence
• Finding: Each worker embeds the full submitted code directly into prompts and sends it to the LLM, again without explicit disclosure or minimization. Because this is a multi-agent design, the same sensitive code may be replicated across several prompts, increasing data exposure surface, third-party retention risk, and the chance of leaking secrets or regulated content.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The manager sends raw task content directly to an LLM in decompose(), which can expose source code, secrets, proprietary data, or user-supplied content to an external model provider without any disclosure, consent, or minimization. In a code-review skill, this is especially relevant because tasks are likely to contain full code snippets, credentials by mistake, or sensitive internal implementation details.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The integrate() step forwards worker outputs and error messages to the LLM, which can compound data leakage by transmitting derived findings, code excerpts, stack traces, and potentially sensitive failure details to an external service. Because this skill is a multi-agent code review system, aggregated worker results may contain a broader and more sensitive view of the codebase than the original single task, increasing confidentiality risk.

VirusTotal

VirusTotal findings are pending for this skill version.