Nm Scribe Voice Extract

• Outcome: Review
• Latest audit: Jun 19, 2026 at 2:38 PM UTC
• Version:
  1.9.12

Overview

This skill has a coherent writing-voice purpose, but it stores sensitive writing samples and reusable voice profiles persistently without clear consent or retention controls, and includes AI-detectability and imitation-oriented guidance.

Install only if you are comfortable with your writing samples and extracted voice profile being saved locally under ~/.claude/voice-profiles. Use samples you own, avoid private or proprietary text unless necessary, review and delete stored samples after extraction, and do not use the profile to impersonate others or conceal AI involvement where disclosure is expected.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Rogue Agent: Self-Modification, Session Persistence
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands

Findings (8)

Context-Inappropriate Capability

Medium

• Confidence: 95% confidence
• Finding: The skill includes a section explicitly labeled as 'Detection-Neutral' and states that the techniques improve writing 'without increasing AI detectability.' That is not necessary for legitimate voice extraction and adds evasion-oriented guidance that could be used to help generated text avoid detection or scrutiny, expanding the skill beyond benign analysis into operational misuse.

Intent-Code Divergence

Medium

• Confidence: 98% confidence
• Finding: The module promises anonymization but the only provided directory-mode implementation performs a raw file copy, which can preserve names, dates, URLs, metadata, and identifying text. In a voice-extraction skill, users are likely to provide sensitive writing samples, so this mismatch can create a false sense of privacy and lead to unintended retention and downstream processing of identifying information.

Vague Triggers

Medium

• Confidence: 84% confidence
• Finding: The trigger list includes very broad terms such as 'voice,' 'writing,' and 'style-transfer,' which are likely to match many unrelated conversations. Overbroad invocation can cause the skill to activate unexpectedly in contexts involving sensitive text, causing unnecessary collection, processing, or persistence of user writing samples.

Missing User Warnings

Medium

• Confidence: 91% confidence
• Finding: The skill instructs the agent to create a persistent profile directory under the user's home directory and copy writing samples into it, but it does not require a clear upfront warning or explicit consent for local data retention. Because the samples may contain personal, sensitive, or proprietary text, silent persistence increases privacy and data-handling risk.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: Directory mode copies user-provided files into a persistent profile directory under the home folder without clearly warning that local duplicates will be created and retained. Because the skill is specifically designed to ingest personal writing for style analysis, those samples may contain sensitive personal, professional, or proprietary content, making silent persistence a meaningful privacy risk.

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: Interactive mode instructs the user to paste writing samples and save them to disk, but it does not disclose that potentially sensitive pasted content will be retained locally. In this skill context, users may paste emails, drafts, journals, or work product for voice modeling, so undisclosed storage increases the risk of privacy breaches, over-retention, and accidental later exposure.

Ssd 2

Medium

• Confidence: 94% confidence
• Finding: The output guidance goes beyond style analysis and provides imitation-oriented 'craft rules' while explicitly tying them to avoiding increased detectability. In context, this facilitates generation of text that mimics a person's writing style in a way that may be harder to identify as synthetic, which materially raises impersonation and abuse risk.

Session Persistence

Medium

• Category: Rogue Agent

Content

# Register Creation Module

Create voice registers from extraction output.

## Default Register

• Confidence: 91% confidence
• Finding: Create voice registers from extraction output. ## Default Register After extraction completes, create the default register at `~/.claude

VirusTotal

63/63 vendors flagged this skill as clean.