ClawHub

Iceberg Excavation

Security checks across malware telemetry and agentic risk

Overview

This is an instruction-only emotional self-reflection skill with no executable code or external access, but users should be aware it can make sensitive psychological inferences and may remember analysis-depth preferences.

Install this only if you want reflective emotional-pattern analysis, not therapy or diagnosis. Avoid sharing crisis, trauma, or highly private details unless you are comfortable, and tell the agent not to save memories unless you explicitly want preferences such as shallow versus deep analysis remembered.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Vague Triggers

Medium
Confidence
92% confidence
Finding
The trigger conditions are broad enough to activate on ordinary emotional or reflective conversation, causing the skill to perform unsolicited deep psychological inference. In this context, the skill is designed to elicit and structure highly sensitive mental-health-adjacent content, so accidental invocation materially increases privacy and safety risk.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The skill handles sensitive emotional and psychological disclosures but does not clearly warn users that their inputs may be inferred into structured psychological interpretations and potentially recorded in memory or logs. Because the workflow includes error logging and preference storage, the absence of explicit notice and consent increases the risk of unexpected retention and misuse of highly sensitive personal data.

Natural-Language Policy Violations

Medium
Confidence
78% confidence
Finding
The skill hard-codes a China-specific cultural framing ('面子', '权威') without user choice, which can steer interpretation of emotional issues through a culturally narrow lens. In a psychological analysis context, this can produce biased or inaccurate inferences that feel authoritative, increasing the risk of harmful mischaracterization for users from other backgrounds.

Ssd 3

Medium
Confidence
97% confidence
Finding
The skill explicitly instructs recording the user's resistance/preferences to memory during sensitive emotional analysis, but provides no consent, minimization, retention, or access controls. Storing mental-health-adjacent preferences in persistent memory can create a durable profile of vulnerability and may expose sensitive information across future interactions or systems.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal