ARIS: Autonomous ML Research through Lightweight Markdown-Driven Workflows

ARIS (Auto-Research-In-Sleep) introduces a framework-agnostic approach to autonomous machine learning research, utilizing Markdown-based skills to facilitate cross-model review loops and experiment automation.

Overview of ARIS (Auto-Research-In-Sleep)

Developed by wanshuiyin, ARIS is designed to streamline the iterative process of machine learning research. Unlike heavy orchestration frameworks, ARIS focuses on a "lightweight" philosophy, employing Markdown-only skills to guide autonomous agents through the complex lifecycle of ML discovery and implementation.

Core Technical Capabilities

The system is engineered to handle several critical stages of the research pipeline without requiring a proprietary ecosystem, ensuring flexibility for the researcher. Key functionalities include:

• Cross-Model Review Loops: Implementing iterative feedback cycles where different models can review, critique, and refine code and theoretical approaches.
• Idea Discovery: Automating the exploration of new hypotheses and research directions.
• Experiment Automation: Reducing manual overhead by automating the execution and tracking of ML experiments.

Interoperability and Integration

A primary advantage of ARIS is its lack of vendor lock-in. Because it relies on a Markdown-based skill set rather than a rigid framework, it is compatible with a wide array of Large Language Model (LLM) agents. Supported integrations include:

• Claude Code
• Codex
• OpenClaw
• Generic LLM agents capable of interpreting Markdown instructions

Architectural Philosophy

By avoiding complex framework dependencies, ARIS ensures that the research logic remains transparent and portable. This approach allows researchers to swap underlying models or agents without rewriting the core research logic, focusing instead on the "skills" defined in Markdown.

Note: As the provided source is a repository summary, detailed implementation specifics regarding the exact Markdown syntax and specific automation triggers are not available.