ClawHub

ClawCoach Core

Security checks across malware telemetry and agentic risk

Overview

This is a coherent health-coaching skill, but users should review it because it stores sensitive diet data and makes a privacy claim that conflicts with its API-backed LLM requirement.

Install only if you are comfortable storing diet, profile, and meal-history data under ~/.clawcoach/ and potentially having that data processed by the configured LLM backend. Review the companion setup and food-analysis skills before relying on the full workflow, and avoid entering sensitive medical or eating-disorder information unless the external processing and retention behavior are clear to you.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

Intent-Code Divergence

Medium
Confidence
92% confidence
Finding
The skill claims that no data leaves the machine, but it explicitly depends on an LLM backend using an external API key for generating coaching responses. That creates a privacy transparency issue: user health, nutrition, and profile data may be sent to a remote service despite documentation asserting otherwise, which can mislead users into sharing sensitive information under false assumptions.

Vague Triggers

Medium
Confidence
78% confidence
Finding
Using a generic trigger like "Help" can cause accidental invocation during ordinary assistant interactions, leading the skill to activate outside the user's intended context. In a skill that reads local health/profile files and provides persona-driven responses, unintended activation may expose sensitive summaries or alter the conversation flow unexpectedly.

Ssd 3

Medium
Confidence
87% confidence
Finding
The persona explicitly directs the model to use the user's actual data in roasts, which increases the chance of unnecessarily repeating sensitive health and nutrition information back to the user in a more vivid or memorable way. In a health-coaching context, this can amplify privacy harm, oversharing, and emotional risk, especially if responses surface exact metrics or patterns more aggressively than needed.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal