The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill exposes multiple tools that allow for arbitrary code execution within the context of target processes.
The inject_dll tool allows the agent to load and execute external libraries into a running process via a provided file path.
The execute_code tool enables the execution of raw assembly or machine code strings at specific memory addresses.
The auto_assemble tool permits the agent to write and execute scripts that can hook functions, allocate memory, and modify program logic.
[COMMAND_EXECUTION]: The configuration section explicitly mentions an environment variable CE_MCP_ALLOW_SHELL. When set to 1, this allows for arbitrary shell command execution, which is noted as dangerous.
[DATA_EXFILTRATION]: The skill provides comprehensive tools for reading arbitrary process memory, including read_memory, read_string, and read_pointer_chain. This capability can be used to harvest sensitive information (credentials, session tokens, personal data) from other applications running on the system. The save_table tool further enables writing this data to the local filesystem.
[PRIVILEGE_ESCALATION]: The skill instructions encourage the use of Cheat Engine's DBVM (Ring -1 hypervisor) for advanced debugging and anti-cheat evasion. This grants the agent access to system-level debugging capabilities that operate below the operating system's standard security controls.
[DYNAMIC_EXECUTION]: The skill facilitates runtime code generation and execution through the auto_assemble and execute_code tools, which compile and run machine code within target processes dynamically.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process unstructured data from external processes, which could contain malicious instructions.
Ingestion points: Tools such as read_string, read_memory, disassemble, and get_rtti_classname fetch data from target process memory into the agent's context.
Capability inventory: The skill possesses high-impact capabilities including write_memory, execute_code, inject_dll, and save_table (documented in SKILL.md).
Boundary markers: There are no documented boundary markers or instructions to the agent to ignore potentially malicious content found within the memory it reads.
Sanitization: The skill does not describe any sanitization or validation of the memory contents before they are interpreted by the AI agent.

cheatengine-mcp-bridge