The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides specific command sequences to exploit multiple services, including Redis, Rsync, and Hadoop YARN.
[REMOTE_CODE_EXECUTION]: Detailed instructions for achieving remote code execution via FastCGI (PHP-FPM), AJP (Ghostcat), and YARN Resource Manager targeting external systems.
[PRIVILEGE_ESCALATION]: Provides methods to write SSH public keys to the root user's directory and inject malicious cron jobs to execute commands with elevated privileges on target infrastructure.
[PERSISTENCE]: Instructions for establishing persistent access by modifying crontabs and SSH authorized_keys files.
[PROMPT_INJECTION]: The 'AI LOAD INSTRUCTION' encourages the model to adopt an 'Expert Attack Playbook' persona, which may override general safety constraints regarding the generation of offensive exploit code.
[EXTERNAL_DOWNLOADS]: References multiple external exploitation tools (e.g., ajpShooter.py, redis-rogue-server.py, gopherus.py) and instructs the agent to utilize them for penetration testing tasks.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process output from scanning tools (like nmap) and execute complex exploitation logic based on that untrusted external data.
Ingestion points: Service responses and scan results from TARGET hosts during discovery phases.
Boundary markers: None present in the instructions to separate data from command logic.
Capability inventory: Includes remote command execution, file system modification, and network requests via tools like curl and redis-cli.
Sanitization: No sanitization or validation logic is defined for service responses before they are interpolated into exploitation commands.

unauthorized-access-common-services