ClawHub
Back to skill

Security audit

aurum-brain

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed reasoning-quality helper with an optional local logger, but users should avoid putting secrets into that logger.

Install only if you want a meta reasoning/checklist workflow. Do not pass API keys, passwords, private prompts, regulated data, or other sensitive text to scripts/reasoning_log.py unless you first redact it, because the helper stores its inputs in a persistent local JSONL file.

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

Medium
Confidence
96% confidence
Finding
The module docstring promises that no secrets are ever written, but the script directly logs arbitrary user-supplied task, plan, and self-check text to a local JSONL file without any filtering or redaction. In a reasoning/logging skill, those fields can easily contain credentials, internal prompts, sensitive business data, or regulated information, so the misleading safety claim can cause unsafe operational use and unintentional data exposure.

Static analysis

No suspicious patterns detected.