skill-security-reviewer
por ninjagpt
Git & GitHubResumen
<!-- Skill Security Reviewer | Version 3.0.0 | Author: chris@zast.ai -->.
Descripción
Qué hace esta skill
Performs static security analysis on Claude skills to detect malicious behavior before or after installation. Covers 94 checks across obfuscation techniques (Base64, XOR, string splitting, dynamic code generation) and threat categories including credential theft, reverse shells, persistence mechanisms, and prompt injection. Operates read-only and outputs a scored markdown report.
It decodes and reconstructs obfuscated content before threat-checking, catching attacks that simple keyword grep or manual review would miss.
Cuándo usarla
- Vetting a downloaded skill from an unknown author before installing it
- Auditing all skills in a shared team Claude environment
- Investigating a skill after noticing unexpected file access or network activity
- Checking if a skill hides malicious commands behind Base64 or XOR encoding
- Running a security review before approving a skill for an organizational deployment
Flujo de trabajo de ejemplo
User asks: /skill-security-reviewer suspicious-data-helper
- 1
Locates skill files under ~/.claude/skills/suspicious-data-helper/
- 2
Reads all files and calculates Shannon entropy per content block
- 3
Detects encoding patterns (Base64, hex, Unicode escapes) and attempts decoding, recursively handling multi-layer nesting
- 4
Runs 53 threat checks on both original and decoded content covering theft, execution, persistence, exfiltration, and prompt injection
- 5
Calculates risk score with obfuscation bonuses and writes a detailed report
Report saved to ./suspicious-data-helper-review-report/report-20260316-143022.md with risk score 85/100 rated High Risk, listing decoded Base64 string containing curl|bash command as primary finding
Instalación
clawhub install ninjagpt/skill-security-reviewerEjecuta el comando de la CLI o pega el prompt en Claude Code, Cursor, Codex o el agente de programación que uses.