This is an experimental research project designed to act as an unconstrained, highly autonomous AI Agent. It is not currently recommended for production workloads.

This framework provides the LLM (Large Language Model) with direct, interactive access to your local physical OS via tools like run_bash_command and edit_file.

The models governing this framework (like GPT-4-class models) can and will autonomously generate and execute complex, multi-stage terminal operations. Running this agent on a system with highly sensitive data or production-critical state is not advised.

• Sandbox Environment: ALWAYS run this agent in an isolated environment (e.g., Docker container, Virtual Machine) if you expand the agent's autonomous limits.
• Human-In-The-Loop (HITL): Keep the builtin Confirm prompts enabled in core/tools.py. Do not bypass them unless the run is strictly sandboxed.
• Tool Tracing: Monitor output closely.

If you've noticed abnormal prompt injection vulnerabilities or accidental systemic execution behaviors during agent testing, please report it via a GitHub issue directly rather than via email. We track all agent-escape payloads openly so the community can patch prompt architectures.