Summary

Sibyl Research System is a fully autonomous AI scientist designed to automate the entire machine learning research lifecycle, from initial idea generation to conference-ready paper publication, with zero human intervention. It targets researchers, engineers, and power users seeking to accelerate scientific discovery and streamline complex research workflows. The system's core benefit lies in its ability to autonomously iterate, refine, and even self-evolve, continuously improving its research capabilities and output quality.

How It Works

Sibyl is built natively on Claude Code, leveraging its agent ecosystem, skills, plugins, and multi-agent teams. It operates on a dual-loop architecture: an inner loop for research iteration (literature review, hypothesis generation, experiment planning and execution, paper writing, and peer review) and an outer loop for system self-evolution. The self-evolution mechanism analyzes completed research iterations, extracts lessons learned across 8 categories, and automatically updates agent prompts and scheduling strategies, enabling the system to improve its own research process over time. This approach allows for autonomous, multi-dimensional iteration and continuous system enhancement.

Quick Start & Requirements

The recommended setup involves cloning the repository, opening it in Claude Code, and instructing Claude to configure everything. Alternatively, a manual setup script (setup.sh) is provided.

• Primary Install/Run:
  • Clone repo: git clone https://github.com/Sibyl-Research-Team/sibyl-research-system.git
  • Navigate: cd sibyl-research-system
  • Run setup: chmod +x setup.sh && ./setup.sh
  • Launch Claude Code with permissions: claude --plugin-dir ./plugin --dangerously-skip-permissions
  • Initialize: /sibyl-research:init (in Claude Code)
  • Start project: /sibyl-research:start spec.md (from workspace root)
• Prerequisites: Python 3.12+, Node.js 18+, Claude Code CLI, GPU server with SSH access, ANTHROPIC_API_KEY environment variable, CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 environment variable, tmux (strongly recommended).
• Links: Full Setup Guide, Configuration, MCP Servers, SSH & GPU Setup.

Highlighted Details

• 19-Stage Research Pipeline: Automates end-to-end research, from literature survey and multi-agent idea debate to GPU-parallel experiment execution and multi-agent paper writing/peer review.
• Multi-Agent Collaboration: Utilizes 20+ specialized AI agents, including 6-agent teams for idea generation, result analysis, and paper critique.
• Self-Evolving System: Learns from research iterations, updates agent prompts and strategies, and accumulates institutional knowledge across projects.
• Self-Healing System: Autonomously detects and fixes runtime errors in real-time, generating regression tests to prevent recurrence.
• Claude Code Native: Deep integration with Claude Code's agent ecosystem, MCP servers, and multi-model collaboration (e.g., Claude Opus/Sonnet + GPT-5.4).
• GPU-Parallel Scheduling: Optimizes GPU utilization through topological sorting and dynamic dispatch of experiment tasks.

Maintenance & Community

The README indicates active development with recent updates in March 2026. Specific details on notable contributors, sponsorships, or community channels (like Discord/Slack) are not provided.

Licensing & Compatibility

The project is released under the MIT License, which generally permits commercial use and modification. No specific compatibility restrictions for closed-source linking are mentioned.

Limitations & Caveats

Execution requires a dedicated GPU server with SSH access for experiment running. The --dangerously-skip-permissions flag, while necessary for full autonomy, bypasses security checks and requires careful environment management to mitigate risks. The setup process involves configuring multiple components (Claude Code, MCP servers, SSH/GPU access) and may be complex for users unfamiliar with these technologies.Summary

Sibyl Research System is a fully autonomous AI scientist designed to automate the entire machine learning research lifecycle, from initial idea generation to conference-ready paper publication, with zero human intervention. It targets researchers, engineers, and power users seeking to accelerate scientific discovery and streamline complex research workflows. The system's core benefit lies in its ability to autonomously iterate, refine, and even self-evolve, continuously improving its research capabilities and output quality.

How It Works

Sibyl is built natively on Claude Code, leveraging its agent ecosystem, skills, plugins, and multi-agent teams. It operates on a dual-loop architecture: an inner loop for research iteration (literature review, hypothesis generation, experiment planning and execution, paper writing, and peer review) and an outer loop for system self-evolution. The self-evolution mechanism analyzes completed research iterations, extracts lessons learned across 8 categories, and automatically updates agent prompts and scheduling strategies, enabling the system to improve its own research process over time. This approach allows for autonomous, multi-dimensional iteration and continuous system enhancement.

Quick Start & Requirements

The recommended setup involves cloning the repository, opening it in Claude Code, and instructing Claude to configure everything. Alternatively, a manual setup script (setup.sh) is provided.

• Primary Install/Run:
  • Clone repo: git clone https://github.com/Sibyl-Research-Team/sibyl-research-system.git
  • Navigate: cd sibyl-research-system
  • Run setup: chmod +x setup.sh && ./setup.sh
  • Launch Claude Code with permissions: claude --plugin-dir ./plugin --dangerously-skip-permissions
  • Initialize: /sibyl-research:init (in Claude Code)
  • Start project: /sibyl-research:start spec.md (from workspace root)
• Prerequisites: Python 3.12+, Node.js 18+, Claude Code CLI, GPU server with SSH access, ANTHROPIC_API_KEY environment variable, CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 environment variable, tmux (strongly recommended).
• Links: Full Setup Guide, Configuration, MCP Servers, SSH & GPU Setup.

Highlighted Details

• 19-Stage Research Pipeline: Automates end-to-end research, from literature survey and multi-agent idea debate to GPU-parallel experiment execution and multi-agent paper writing/peer review.
• Multi-Agent Collaboration: Utilizes 20+ specialized AI agents, including 6-agent teams for idea generation, result analysis, and paper critique.
• Self-Evolving System: Learns from research iterations, updates agent prompts and strategies, and accumulates institutional knowledge across projects.
• Self-Healing System: Autonomously detects and fixes runtime errors in real-time, generating regression tests to prevent recurrence.
• Claude Code Native: Deep integration with Claude Code's agent ecosystem, MCP servers, and multi-model collaboration (e.g., Claude Opus/Sonnet + GPT-5.4).
• GPU-Parallel Scheduling: Optimizes GPU utilization through topological sorting and dynamic dispatch of experiment tasks.

Maintenance & Community

The README indicates active development with recent updates in March 2026. Specific details on notable contributors, sponsorships, or community channels (like Discord/Slack) are not provided.

Licensing & Compatibility

The project is released under the MIT License, which generally permits commercial use and modification. No specific compatibility restrictions for closed-source linking are mentioned.

Limitations & Caveats

Execution requires a dedicated GPU server with SSH access for experiment running. The --dangerously-skip-permissions flag, while necessary for full autonomy, bypasses security checks and requires careful environment management to mitigate risks. The setup process involves configuring multiple components (Claude Code, MCP servers, SSH/GPU access) and may be complex for users unfamiliar with these technologies.