Imprint

• Outcome: Pass
• Latest audit: May 29, 2026 at 8:26 PM UTC
• Version:
  1.0.1

Overview

This skill intentionally builds a local profile of the user's work style, which is privacy-sensitive but clearly disclosed and aligned with its purpose.

Install only if you want your agent to build a local behavioral profile over time. Review and periodically delete or reset `imprint/operator-model.json` and `imprint/observations/`, and avoid using it in shared workspaces unless everyone understands the profiling behavior.

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 skill explicitly instructs the agent to load a persistent operator profile at session start, silently adapt behavior, and write updated behavioral data at session end without a clear consent or notice flow. Even if the author claims the data is local and derived, this is still persistent behavioral profiling across sessions, which creates privacy, transparency, and potential misuse risks if the workspace is shared, compromised, or repurposed by other components.

Natural-Language Policy Violations

Low

• Confidence: 79% confidence
• Finding: The skill promotes matching the operator's communication style 'naturally' and elsewhere says to apply traits without announcing it, which encourages covert personalization without user opt-in. While not as severe as persistent profiling, silent style mirroring can mislead users about the basis for responses and may cross policy boundaries around unannounced behavioral adaptation, especially when paired with long-term operator modeling.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The skill explicitly instructs the agent to load and update a persistent operator model across sessions, but it does not require clear upfront notice and consent before ongoing behavioral profiling begins. Even though the document includes privacy-oriented constraints later, the collection and persistence of inferred behavioral data without explicit opt-in can create non-obvious surveillance, profiling, and trust risks.

Natural-Language Policy Violations

Low

• Confidence: 78% confidence
• Finding: The skill tells the agent to automatically match communication style before every response based on an inferred model, which can cause silent behavioral adaptation without the operator's explicit preference. In context this is less dangerous than data exfiltration or privilege abuse, but it can still produce manipulative or unwanted mirroring, especially when derived from persistent profiling.

VirusTotal

66/66 vendors flagged this skill as clean.