Reflect Critique Revise

• Outcome: Pass
• Latest audit: May 29, 2026 at 8:41 PM UTC
• Version:
  1.0.0

Overview

This is a coherent code-review helper that uses an LLM to critique and revise supplied code, with no evidence of hidden persistence, credential access, or destructive behavior.

Install if you are comfortable with automatic code-review triggers and with the reviewed code being sent to the configured LLM endpoint. Use a trusted local or approved endpoint, avoid drafts containing secrets, and narrow activation rules if you do not want it invoked by generic review wording.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• 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

Findings (3)

Vague Triggers

Medium

• Confidence: 93% confidence
• Finding: The trigger list includes very common phrases like "review," "double check," and a broad post-condition of "after any code generation producing > 30 lines," which can cause the skill to activate in many ordinary coding interactions. Unintended activation increases LLM calls, broadens exposure of generated code and user context to this skill, and can create unpredictable orchestration behavior or prompt-surface expansion.

Vague Triggers

Medium

• Confidence: 89% confidence
• Finding: The standalone invocation rule allows activation whenever a user says "review this code" or pastes code with "is this right?", which is ambiguous and likely to match routine conversation. In a code-orchestration setting, this can cause unnecessary processing of sensitive code, accidental recursion/chaining with other skills, and excessive token/tool consumption without clear user intent.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: This code transmits the user-supplied task and draft code to a configurable external LLM endpoint, with a default of plain HTTP to localhost and no explicit consent, warning, redaction, or trust validation. In a code-review skill, drafts may contain proprietary source, credentials, tokens, or other sensitive material, so silent exfiltration to an endpoint controlled by the environment or local network is a real confidentiality risk.

VirusTotal

67/67 vendors flagged this skill as clean.