Claude Code My Workflow
A ready-to-fork Claude Code template for academics — Beamer/Quarto slides, R data analysis, and manuscript review, run through 18 specialist agents and quality gates.
Last reviewed by the paperbanana team on Jul 13, 2026
Install
git clone https://github.com/YOUR_USERNAME/claude-code-my-workflow.git my-project
cd my-project
./scripts/validate-setup.shWhat is Claude Code My Workflow?
Claude Code My Workflow is a fork-and-adapt Claude Code template — extracted from a production PhD course — for AI-assisted academic work: lecture slides in Beamer/Quarto, research papers, R-based data analysis, and replication packages. Rather than a single skill, it is an entire repository of 18 specialist agents, 52 skills, 32 rules, and 7 hooks wired together so that you describe a goal ("build lecture slides on X", "review this paper") and Claude plans the approach, invokes the matching skill, and runs an implement → verify → review → fix → re-verify loop internally before returning a result.
The defining mechanism is what the README calls "goal-first, gate-enforced" orchestration: instead of a single AI pass, work loops under enforced quality gates — 80 to commit, 90 for a PR, 95 as an aspirational excellence bar — backed by a real pre-commit hook (installed once via ./scripts/install-hooks.sh) so bypassing the invoking skill no longer bypasses the review. An adversarial critic/fixer pair reviews Beamer and Quarto decks and loops until a pass surfaces no new issue (a 5-round cap is a fallback, not the primary stop condition), and a Chain-of-Verification pass (/verify-claims) uses a forked verifier that has never seen the draft to catch fabricated citations or numerical contradictions before /commit.
The template ships discipline profiles preloaded for economics (top-5 journal conventions, R/causal-inference conventions) and political science (APSR/AJPS/JOP), and extends to other fields via journal profiles and paper-type cards. Coverage spans the full research program rather than one document: lecture creation, literature review, manuscript peer-review simulation (including reviewer-disposition variance reporting), Monte Carlo simulation studies, R package release gates, replication-package assembly, grant proposals, syllabi, and even inbox/calendar triage — all governed by the same CLAUDE.md, rules, and quality gates.
Core capabilities
18 specialist review agents
Narrow agents each check one dimension — proofreader, slide-auditor, pedagogy-reviewer, r-reviewer, tikz-reviewer, domain-referee, methods-referee, editor — rather than one generalist reviewer covering everything at once.
Adversarial critic/fixer QA loop
A critic agent reads both Beamer and Quarto output and produces harsh findings; a fixer agent implements exactly what the critic found. They loop until a round surfaces no new issue, catching errors single-pass review misses.
Quality gates enforced by a real pre-commit hook
Scores below 80 halt /commit; 90 is the PR threshold; 95 is aspirational excellence. A version-controlled pre-commit hook re-runs surface-sync and quality checks on every commit, plus a git-guardrails hook blocking destructive git operations.
Verification and reproducibility checks
/verify-claims runs Chain-of-Verification with a forked verifier that cannot self-confirm; /audit-reproducibility cross-checks every numeric claim in the manuscript against the script output that produced it, tracked in a per-paper passport.yaml.
Full research-program skill set
52 skills cover lecture creation (/create-lecture), literature review (/lit-review), manuscript peer-review simulation (/review-paper --peer, with --variance N disposition sampling), R data analysis (/data-analysis), Monte Carlo studies (/simulation-study), and replication packaging.
Session continuity across compaction
Plans, specs, and session logs survive auto-compression via a PreCompact hook snapshot; MEMORY.md accumulates cross-session learnings, and /compress-session distills a session into decisions, next actions, and discarded-as-noise before forced compaction.
What you can use it for
Building a lecture deck with multi-agent review
Run /create-lecture, then /slide-excellence to fan out proofreader, slide-auditor, pedagogy-reviewer, and tikz-reviewer in parallel before the deck is considered done.
Simulating peer review before submission
Use /review-paper --peer [journal] to run a desk-review-plus-referee pipeline with the editor, domain-referee, and methods-referee agents, optionally with --variance N to see how much the decision would vary by reviewer disposition.
Running a full R-based empirical analysis
/data-analysis drives an end-to-end R pipeline with reproducibility conventions, output verification, and publication-ready tables/figures as a single orchestrated task.
Assembling a replication package for submission
/replication-package composes a DCAS/openICPSR-compliant bundle — README, dataset manifest, Table/Figure-to-script mapping — and blocks on a failing /audit-reproducibility check.
Translating a Beamer deck to Quarto for the web
/translate-to-quarto runs an 11-phase translation with a beamer-translator agent and a quarto-critic/quarto-fixer adversarial loop to match the Beamer benchmark.
How to get started
- 1
Fork and clone the repo
Fork on GitHub, then clone your fork and run ./scripts/validate-setup.sh, which reports exactly what tools (XeLaTeX, Quarto, R, git) are missing with install links.
- 2
Start Claude Code and paste the starter prompt
Run `claude` (or open the Claude Code panel in VS Code) and paste the starter prompt from the workflow guide, describing your project in 2-3 sentences so Claude adapts CLAUDE.md.
- 3
Verify the environment with the HelloWorld demos
Run /compile-latex HelloWorld and /deploy HelloWorld inside Claude to confirm Beamer and Quarto both compile before deleting the demo files and starting real work.
- 4
Install the quality-gate hook
Run ./scripts/install-hooks.sh once to install the real pre-commit hook so every /commit is gated by the surface-sync and ≥80 quality checks, not just when the skill is invoked directly.
How it compares to similar skills
This template is a heavyweight, forkable research-program hub rather than a single drop-in skill — it is closest in spirit to Academic Research Skills but built around R/econometrics conventions and Beamer/Quarto slides specifically. Lighter or more narrowly scoped alternatives in this directory:
Academic Research Skills
Pick Academic Research Skills for a discipline-agnostic research → write → review pipeline without the R/econometrics-specific tooling and Beamer/Quarto slide focus.
Academic PPTX Skill
Pick Academic PPTX Skill if you only need action-title, argument-structured PowerPoint slides rather than a full forked repository with R analysis and manuscript review.
LaTeX Document Skill
Pick LaTeX Document Skill if you want a single Claude Code skill for general LaTeX document generation (theses, posters, resumes) without adopting an entire forked research-program repo.