ClawHub
Back to skill

Security audit

Video Color Grade

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed local video color-grading skill that runs media-processing commands and writes project outputs, with no hidden network, credential, persistence, or destructive behavior found.

Install only if you are comfortable letting the agent run ffmpeg/Python and read/write files in your video project. Check input and output paths before running because the scripts use overwrite flags for generated media outputs.

SkillSpector

Vulnerability Patterns
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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)

Lp3

Medium
Category
MCP Least Privilege
Confidence
95% confidence
Finding
The skill instructs the agent to execute shell commands and write multiple files, but it declares no permissions. That mismatch is a real security issue because it prevents proper consent, policy enforcement, and sandboxing decisions by the host, and could enable unintended command execution or filesystem modification if the skill is invoked in a more privileged environment.

Static analysis

No suspicious patterns detected.