ClawHub
Back to skill

Security audit

Universal Game Primitives

Security checks for vulnerabilities and agentic risk

Overview

This is a markdown-only game architecture helper with examples, and I found no hidden execution, persistence, credential access, or unsafe authority.

This skill is reasonable to install for game architecture design. Treat the long code snippets as illustrative patterns and review them before copying into a project, especially the NPC wattitude threshold example, which appears to need correction.

A.I.G

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Anti-Refusal Statement

High
Category
Anti-Refusal
Content
## When NOT to use it

- Don't invent new primitives (keep the vocabulary stable)
- Don't debate engine religion (Unity vs Unreal vs custom). Translate, don't preach.
- Don't skip the concrete artifact: **every use must end in a diagram, table, or pseudocode**

## The DSL (definitions)
Confidence
85% confidence
Finding
Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Intent-Code Divergence

Medium
Confidence
96% confidence
Finding
At L053-L054, the documentation says intent emission has "no direct side-effects here," implying a pure event-production step. But the full implementation later mutates NPC state throughout the tick and calls `executeNPCAction(npc, action, world, dt)` at L492, which performs direct execution rather than just event emission, so the documented intent and actual behavior diverge.

Intent-Code Divergence

Medium
Confidence
99% confidence
Finding
The comment and type at L266-L267 define four behavioral bands: COOL, WARM, HOT, and CRITICAL. However, the implementation at L269-L273 returns `CRITICAL` for all values >= 50, making the HOT band unreachable and contradicting the stated band model used by the surrounding documentation and behavior table.

Missing User Warnings

Low
Confidence
82% confidence
Finding
This markdown file documents that at HOT+ wattitude the NPC's memory fields are actively cleared, including known threats and conversation partner state. The surrounding description presents this as implementation behavior, but does not include any warning that the skill can intentionally discard in-memory state or alter conversational continuity.

Static analysis

No suspicious patterns detected.