Summary

This repository provides a suite of tools, templates, and philosophies designed to enhance AI-assisted workflows for coding, teaching, and presentations, primarily targeting researchers and educators. It emphasizes rigorous auditing, cross-language replication, and clear communication strategies to improve the reliability and understanding of AI-generated outputs, positioning the AI as a collaborative "thinking partner."

How It Works

The core methodology treats AI as a collaborator, utilizing markdown for persistent project context, cross-software replication (R=Stata=Python to 6 decimal places) for verification, and adversarial review via fresh AI instances. Key components include "Referee 2," a systematic five-audit protocol for independent project auditing; "Fletcher," a checklist for interrogating empirical output; the "Rhetoric of Decks" for designing clear Beamer presentations; and a "Split-PDF Skill" for deep-reading long documents by processing them in manageable chunks.

Quick Start & Requirements

To begin, copy the claude/CLAUDE.md project context template to your project root. Subsequent steps involve invoking AI commands such as /referee2, /fletcher, or /split-pdf within an AI coding environment (e.g., Claude Code). Prerequisites include the AI environment and relevant language interpreters (R, Stata, Python) for replication tasks. Detailed guidance is available in workflow.md and specific tool documentation.

Highlighted Details

• Referee 2 Protocol: A rigorous five-audit process (Code, Cross-Language Replication, Directory, Output Automation, Econometrics) executed by an independent AI instance to ensure objectivity.
• Cross-Language Replication: Mandates identical results to 6+ decimal places across R, Stata, and Python implementations to detect subtle bugs.
• Fletcher Audit: A structured six-step interrogation of empirical output to ensure complete understanding and identification of unexpected patterns.
• Rhetoric of Decks: Promotes presentation design principles like "titles as assertions" and "MB/MC equivalence" for effective audience engagement.

Maintenance & Community

Authored by Scott Cunningham, this collection is actively developed with new approaches added periodically. Specific community channels (e.g., Discord, Slack) or a formal roadmap are not detailed in the README.

Licensing & Compatibility

The project is available for free use, with attribution appreciated but not required, indicating a permissive open-source model. It is designed for integration with AI coding assistants and supports cross-language replication (R, Stata, Python) and LaTeX compilation.

Limitations & Caveats

The tools are presented as personal workflows ("Your mileage may vary") and require a dedicated, fresh AI session for the Referee 2 audit to ensure true independence. Effective use depends on adhering to the described philosophy and maintaining a structured project directory.