The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains various prompt injection payloads and strategies (e.g., in aradotso-ai-security.md and aradotso-agent-safety.md). These are explicitly documented as test cases for red-teaming third-party AI applications and auditing 'vibe-coded' apps. They are not intended to override the agent's own instructions but to serve as a functional toolkit for security professionals.
[COMMAND_EXECUTION]: The skill makes extensive use of CLI-based security tools such as ffuf, nuclei, semgrep, dalfox, and zap-cli. A mandatory 'Authorization Gate' defined in references/authorization-disclaimer.md acts as a safety control, requiring the agent to obtain explicit user confirmation ('yes', 'ack') before executing any commands that interact with live network targets.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for downloading and installing numerous open-source security tools via git clone, pip, and npm (e.g., osv-scanner, trufflehog, gitleaks). These refer to well-known repositories and packages within the cybersecurity community.
[REMOTE_CODE_EXECUTION]: The skill uses npx -y to execute MCP servers like security-detections-mcp and agent-security-scanner. This facilitates the dynamic loading of security-focused capabilities and is standard for the specialized 'router/dispatcher' execution model described in the skill.

cybersecurity