ClawHub
Back to skill

Security audit

Noise Maker

Security checks across malware telemetry and agentic risk

Overview

This is a transparent planning document for an alert gateway, not runnable code, and its sensitive operational ideas are disclosed and fit the monitoring purpose.

Treat this as an architecture/runbook, not a ready-to-run monitoring skill. Before any future implementation, verify confirmation prompts or authorization for closing incidents and changing rules, restrict database and Telegram credentials, define raw alert retention, and review permissions on any local buffer, cache, or startup hook.

SkillSpector

By NVIDIA
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

Medium
Confidence
83% confidence
Finding
The skill explicitly defines operational scripts such as 'close_incident.sh' and 'toggle_rule.sh' that can change monitoring state, but the document provides no confirmation, authorization, or safety expectations around their use. In an agent-skill context, terse side-effecting tools without guardrails increase the risk of accidental incident closure or disabling alerting rules, which can suppress detection and delay response.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal