ClawHub
Back to skill

Security audit

Don't Starve Guide

Security checks across malware telemetry and agentic risk

Overview

This skill is a coherent Don't Starve guide that keeps a limited local gameplay profile, with no evidence of exfiltration, destructive behavior, or unrelated authority.

Install only if you are comfortable with the skill keeping local gameplay facts such as version, progress, preferred characters, and preferences. Users who want isolation should set DONTSTARVE_MEMORY_DIR to a dedicated directory and review or remove ~/.config/dont-starve-skill/survivor-profile.json if they do not want retained profile data.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Lp3

Medium
Category
MCP Least Privilege
Confidence
93% confidence
Finding
The skill invokes local code and uses environment variables plus file reads/writes, but it does not declare those capabilities or obtain explicit user consent. Hidden capability use is dangerous because users and the hosting platform cannot accurately assess what the skill will access or modify, especially when it writes persistent data under the user's home/config directory.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding
The skill is presented as a gameplay/lore guide, but its instructions add persistent state management that reads, writes, merges, and migrates local user profile data. This mismatch undermines informed consent and trust: a user invoking a game guide would not reasonably expect background data persistence and profile mutation, which can lead to undisclosed collection of behavioral data over time.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The skill explicitly instructs the agent to persist user-derived facts after answering, but it provides no user-facing warning at the time of collection or storage. Silent persistence is risky because even 'structured facts' about preferences, progress, and play style accumulate into a profile the user did not knowingly authorize, creating privacy and trust concerns.

Natural-Language Policy Violations

Medium
Confidence
95% confidence
Finding
The examples are entirely written in Chinese and present formatting rules that can implicitly steer the skill to answer in Chinese by default, even when the user has not requested that language. This is not a classic security exploit, but it is a genuine policy/interaction-quality issue because it can override user preference, reduce usability, and cause misleading or inaccessible responses for non-Chinese users.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

Detected: suspicious.dynamic_code_execution

Dynamic code execution detected.

Critical
Code
suspicious.dynamic_code_execution
Location
tests/conftest.py:23