ClawHub
Back to skill

Security audit

SATAgent

Security checks for vulnerabilities and agentic risk

Overview

This appears to be a purpose-aligned security scanning skill, but users should run it only on intended repositories and review what files it may read or write.

Install only from the expected publisher/source, run it in a repository you intend to scan, and review any generated reports or SBOM output before sharing. Avoid pointing it at broad home-directory paths unless you are comfortable with local config and dependency metadata being inspected.

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 documentation instructs the agent to invoke local Python scripts, read and write repository artifacts, and operate on arbitrary repo paths, which implies shell, file read, and file write capabilities; it may also process external CVE/manifests, suggesting network-capable workflows. Because these capabilities are not explicitly declared, users and policy systems cannot accurately scope or review what the skill can do, increasing the risk of unintended command execution or data access.

Static analysis

No suspicious patterns detected.