The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines an MCP tool execute_command that allows the agent to run any client command within the IoM environment. This includes high-risk operations such as arbitrary shell execution (shell <command>), process manipulation (ps, kill), and system discovery (whoami, privs).- [REMOTE_CODE_EXECUTION]: The skill includes the execute_lua tool, which enables the execution of arbitrary Lua scripts with access to internal IoM APIs. Furthermore, it documents the mal install command used to download and execute code from a remote community repository.- [DATA_EXFILTRATION]: The framework explicitly supports upload and download commands, allowing for the unauthorized movement of files between the local environment and a remote C2 server.- [PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection by ingesting data from an external C2 infrastructure (implant tasks and server events) which could be used to influence agent behavior.
Ingestion points: Task results and command history retrieved via execute_command and get_history tools.
Boundary markers: None described in the documentation or tool definitions.
Capability inventory: High-privilege actions including command execution, Lua scripting, and file system access.
Sanitization: No mechanisms for validating or escaping external framework output are identified.

iom