lsdefine/GenericAgent

What it solves

GenericAgent is a minimal, self-evolving autonomous agent framework designed to give LLMs system-level control over a local computer. It eliminates the need to preload a vast library of specialized skills by allowing the agent to autonomously discover, execute, and then "crystallize" successful execution paths into reusable skills, reducing token consumption and hallucinations.

How it works

The framework operates on a minimal core of approximately 3,000 lines of code and a ~100-line execution loop. It uses a layered memory system (L0-L4) to store meta-rules, global facts, and task-specific SOPs. The agent interacts with the world through 9 atomic tools (such as code execution, file manipulation, and web scanning).

To expand its capabilities, the agent follows a self-evolution mechanism: it explores a new task, installs necessary dependencies, writes scripts, and verifies the result. Once successful, it saves the process as a "Skill" in its memory, allowing for direct recall and one-line invocation in the future.

Who it’s for

It is for developers and power users who want a lightweight, cross-platform autonomous agent that can control browsers, terminals, and mobile devices (via ADB) without the overhead of massive frameworks or complex deployments.

Highlights

• Self-Evolving Skill Tree: Automatically converts successful task completions into permanent, reusable skills.
• Minimalist Architecture: Core logic is contained in ~3K lines of code, making it highly efficient and easy to deploy.
• Real-Browser Control: Uses TMWebdriver to inject into real browser sessions, preserving logins and bypassing bot detection (e.g., hCaptcha).
• Token Efficient: Operates with a context window of under 30K tokens, significantly lower than many competing agents.
• System-Level Access: Direct control over keyboard, mouse, filesystem, and mobile devices via ADB.