Saltar al contenido
TakoAPI
Volver a skills

skillguard-scanner

por msgnoki

Coding Agents & IDEs
1.1kDescargas según clawskills.sh
2Estrellas
0Vistas
0 me gustaInicia sesión para dar me gusta a esta skill
—Aún sin valoraciones
Tu valoración

Resumen

Security scanner for OpenClaw/ClawHub skills.

Descripción

Qué hace esta skill

SkillGuard scans OpenClaw skills for security threats including reverse shells, credential theft, memory poisoning, and typosquatting. It works on installed skills, individual skill files, or ClawHub skills fetched before installation. There is no official skill vetting process for OpenClaw, and documented campaigns have distributed malware through the ecosystem.

It catches specific attack patterns documented in real ClawHub malware campaigns, including the ClawHavoc vector of embedding download instructions in skill prerequisites, which generic scanners miss.

Cuándo usarla

  • Scan all installed skills before a system audit
  • Check a ClawHub skill for malware before installing it
  • Verify a skill name isn't typosquatting a known skill
  • Review flagged patterns in a skill someone else wrote
  • Automated security check in a skill review workflow

Flujo de trabajo de ejemplo

ENTRADA

User asks: scan the 'git-helper' skill from ClawHub before I install it

AGENTE
  1. 1

    Run scanner with --fetch-clawhub git-helper to pull the skill from ClawHub

  2. 2

    Scanner fetches all skill files and checks each against threat pattern library

  3. 3

    Detects any reverse shell commands, suspicious URLs, or credential access patterns

  4. 4

    Assigns risk score and false positive likelihood to each finding

  5. 5

    Saves full JSON report to data/scan_results.json

SALIDA

Console summary showing risk level (e.g., MEDIUM score 14, one flag for subprocess use with high FP likelihood), plus full JSON report for manual review

Requisitos

  • clawhub CLI (required only for --fetch-clawhub flag)

Instalación

clawhub install msgnoki/skillguard-scanner

Ejecuta el comando de la CLI o pega el prompt en Claude Code, Cursor, Codex o el agente de programación que uses.

Enlaces