This project provides a comprehensive suite of Claude Code skills designed to automate and enhance the entire academic research and writing pipeline, from initial deep research to final publication. Targeting researchers and academics, it offers significant benefits by streamlining complex workflows, improving the quality and integrity of academic papers through rigorous multi-stage review and verification processes, and providing detailed process summaries.

How It Works

The system operates as a 10-stage pipeline orchestrated by Claude Code, leveraging a multi-agent architecture. It integrates deep research capabilities, automated paper writing, and a sophisticated review process involving multiple personas, including a Devil's Advocate. A key differentiator is its emphasis on integrity verification, performing 100% checks on references, data, and claims at critical stages. Socratic coaching is embedded to guide revisions, ensuring a robust and iterative improvement cycle.

Quick Start & Requirements

Installation involves cloning the repository into .claude/skills/ or running it as a standalone project with Claude Code. Key prerequisites include Claude Code (native installer recommended), an Anthropic API key, and a recommended model like Claude Opus 4.6 with a Max plan. For autonomous execution, enabling experimental features like Agent Teams (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1) and claude --dangerously-skip-permissions is advised. The full pipeline is token-intensive, potentially exceeding 200K input + 100K output tokens per run. Installation scripts and API key setup instructions are available.

Highlighted Details

• 10-Stage Pipeline: Covers research, writing, integrity checks, multi-round review (EIC, dynamic reviewers, Devil's Advocate), Socratic revision coaching, and finalization.
• Integrity Verification: Features pre- and post-review validation of references, data, and claims with an anti-hallucination mandate.
• Multi-Perspective Review: Employs a 5-person review team with 0-100 quality rubrics for comprehensive feedback.
• Output Formats: Supports MD, DOCX, and LaTeX (APA 7.0, IEEE, Chicago) outputs, compiled to PDF via Tectonic.
• Collaboration Scoring: Generates a post-pipeline summary with a 6-dimension collaboration quality evaluation.
• Material Passport: Enables provenance tracking for mid-entry into the pipeline.

Maintenance & Community

The project is actively maintained by Cheng-I Wu (吳政宜), with recent updates focusing on integrity verification and intent-based mode activation. The primary community interaction point is the GitHub repository.

Licensing & Compatibility

This work is licensed under CC-BY-NC 4.0. While it allows sharing and adaptation, it strictly prohibits commercial use and requires appropriate attribution.

Limitations & Caveats

The extensive token consumption of the full pipeline necessitates careful budget management. Usage on claude.ai may yield less comprehensive results due to the lack of parallel multi-agent execution. Enabling --dangerously-skip-permissions removes safety nets for autonomous operation. The CC-BY-NC 4.0 license restricts usage to non-commercial purposes.