pedrohcgs/claude-code-my-workflow

What it solves

This project provides a structured framework for AI-assisted academic work, enabling researchers and educators to automate the creation, review, and verification of academic artifacts like lecture slides, research papers, data analysis pipelines, and replication packages. It transforms Claude Code into a "contractor" that can plan, implement, and verify complex academic tasks while maintaining high quality standards through automated gates.

How it works

The system uses a "goal-first, gate-enforced" approach where the user states a goal and the AI handles the orchestration. It employs several key mechanisms:

• Specialized Agents: Instead of a generalist, it uses 18 focused agents (e.g., proofreader, pedagogy-reviewer, r-reviewer) to handle specific dimensions of academic quality.
• Orchestration Runtime: A system that fans out tasks to specialist agents, reduces findings over a shared schema, and loops until issues are resolved.
• ** AdSversarial QA**: A "critic" and "fixer" agent loop until no new issues are found, catching errors that single-pass reviews miss.
• Quality Gates: Artifacts are scored (0-100). Pre-commit hooks and skills (like /commit) enforce minimum quality thresholds (e.g., 80 for commits) to prevent low-quality work from being saved.
• Verification Layers: Dedicated tools for verifying claims (Chain-of-Verification), auditing reproducibility (cross-checking numeric claims against script output), and detecting AI-voice tells (/humanize).
• Context Survival: Uses MEMORY.md and session distillation (/compress-session) to ensure critical decisions and learnings persist across sessions.

Who it’s for

Academic researchers, PhD students, and professors who use LaTeX (Beamer), Quarto, R, and Stata for their research and teaching materials.

Highlights

• Multi-Agent Pipeline: 18 specialized agents for everything from TikZ diagrams to pedagogical review.
• Automated Quality Scoring: Integrated scoring system with hard gates for commits and PRs.
• Reproducibility Audit: Tools to cross-check manuscript claims against actual data output via passport.yaml.
• Simulated Peer Review: The /review-paper --peer skill simulates a journal editor and referees with varying dispositions.
• Academic Tooling Integration: Native support for XeLaTeX, Quarto, R, and Stata (via MCP).
• Monte Carlo Support: Dedicated workflows for simulation studies and R package release checks.