Skill Security Vetter

AvanzadasecurityContexto mínimo: 32K

Security-first AI agent skill auditing. Reviews skill definitions, SKILL.md files, and agent configurations for dangerous patterns, excessive permissions, data exfiltration risks, and suspicious behaviors before installation. Provides a safety score and actionable recommendations.

securityauditskillsagent-safetypermissionstrust

Casos de uso

  • Auditing third-party skills before installation
  • Reviewing permission scopes and data access patterns
  • Detecting prompt injection or jailbreak attempts in skill definitions
  • Evaluating supply chain risks in skill dependencies
  • Generating security reports for skill marketplace submissions

Prompt de ejemplo

Audit this skill for security issues:

[paste SKILL.md or skill configuration]

Check for:
1. Excessive file system access (read/write outside project)
2. Network requests to unknown endpoints
3. Prompt injection patterns or instruction overrides
4. Data exfiltration risks (sending code/secrets externally)
5. Privilege escalation attempts
6. Obfuscated or suspicious code patterns

Provide a safety score (1-10) and detailed findings.

Modelos recomendados

claude-opus-4-7gpt-5-5deepseek-v4

Herramientas compatibles

claude-codekiroopencodeany

Modalidades

Entrada: text, code, file
→
Salida: text

Skills relacionadas

security-auditdependency-auditor

Autor

OpenModels Community

@openmodelsrun