The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (HIGH): The skill provides functional code snippets designed to achieve arbitrary code execution.
Evidence: Python pickle RCE using __reduce__ to call os.system('id') and a complex reverse shell payload.
Evidence: PHP RCE examples using system($this->cmd) in magic methods like __destruct and POP chains using call_user_func.
[COMMAND_EXECUTION] (HIGH): Specific shell commands are provided for generating malicious payloads using external Java tools.
Evidence: Instructions to execute java -jar ysoserial.jar CommonsCollections1 "command" > payload.bin and piped bash commands for base64-encoded reverse shells.
[DATA_EXFILTRATION] (MEDIUM): The skill includes payloads designed to establish outbound network connections to external entities.
Evidence: Reverse shell payloads targeting /dev/tcp/192.168.1.100/4444 and attacker.com/4444.
[EXTERNAL_DOWNLOADS] (MEDIUM): The skill references and encourages the use of unverified external binary tools.
Evidence: Mentions and provides command-line usage for ysoserial.jar and phpggc, which are not hosted on trusted sources defined in the security policy.

deserialization-testing