ClawHub
Back to skill

Security audit

Security Constitution

Security checks for vulnerabilities and agentic risk

Overview

This security skill is mostly coherent, but it asks the agent to act as a broad gatekeeper while documenting a configuration-edit bypass and leaving password, logging, and notification handling under-scoped.

Install only if you want a prompt-level security gate that can interrupt or refuse tasks. Before using it, define who controls security-policy.json, avoid entering a reusable real password in chat, and set explicit rules for what gets logged, who can read logs or notifications, how secrets are redacted, and how long records are kept.

SkillSpector

Vulnerability Patterns
  • 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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (1)

Intent-Code Divergence

High
Confidence
95% confidence
Finding
The skill claims its security rules are absolute and cannot be bypassed, yet the L1 refusal template tells users to 'directly edit the configuration file' to perform blocked actions. That creates a documented bypass path that undermines the stated trust boundary and may let an operator disable protections or reclassify dangerous actions outside the normal approval flow.

Static analysis

No suspicious patterns detected.