ClawHub
Back to skill

Security audit

Prisoner's Dilemma

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent educational skill for applying the prisoner's dilemma framework, with no executable behavior or hidden access.

This skill appears safe to install from a security perspective. Users should be aware that one example discusses wartime violence and that the skill is advisory analysis, not a substitute for legal, policy, or business judgment.

SkillSpector

Vulnerability Patterns
  • 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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (1)

Missing User Warnings

Low
Confidence
82% confidence
Finding
This is a markdown file, so SQP-2 applies to omissions in the skill description. The example repeatedly discusses killing, retaliation, raids, and mutual slaughter, but provides no warning that the material contains graphic or violent subject matter that could affect user expectations or safety context.

Static analysis

No suspicious patterns detected.