pedrohcgs/claude-code-my-workflow

What it solves

This project provides a structured framework for using Claude Code to automate academic research and content creation. It transforms the AI from a simple chat interface into a "contractor" that can plan, implement, and verify the production of academic artifacts like research papers, lecture slides, and data analysis packages, ensuring high quality through rigorous automated gates and specialized agent reviews.

How it works

The system operates on a "goal-first, gate-enforced" philosophy. Instead of manual prompting, the user defines a goal, and the system loops toward it using several mechanisms:

• Specialized Agents: 18 focused agents (e.g., proofreader, slide-auditor, pedagogy-reviewer) handle narrow tasks better than a generalist model.
• Orchestration Runtime: A protocol that fans out work to these agents, reduces findings into a shared schema, and loops until no new issues are found.
• Automated Gates: A suite of scripts (like backtest.sh) and pre-commit hooks that enforce quality scores (e.g., 80+ for commits) and verify technical integrity (e.g., link resolution, skill conformance).
• Adversarial QA: A "critic" and "fixer" agent loop until they converge on a version of the document that is error-free.
• Verification Layers: Specialized skills for verifying claims via Chain-of-Verification, auditing reproducibility by cross-checking numeric claims against script outputs, and detecting AI-voice tells.

Who it’s for

Researchers, PhD students, and academics who need to produce high-quality, verified academic outputs (LaTeX, Quarto, R, Python) while leveraging AI assistance without sacrificing rigor or accuracy.

Highlights

• Contractor Mode: Automates the full cycle of requirement specification, planning, and implementation.
• Reproducibility Audit: Cross-checks manuscript numbers against the actual data scripts to prevent errors.
• Peer Review Simulation: Runs multiple referees with different dispositions to report a decision distribution rather than a single estimate.
• Rigorous Quality Gates: Uses version-controlled hooks and a qualification ledger to ensure that every check is actually effective.
• Context Survival: Implements hooks to save snapshots before auto-compression, ensuring critical decisions are preserved across sessions.