draft-paper

name	draft-paper
description	Draft academic economics papers by dispatching the Writer agent. Handles section drafting, notation protocol, anti-hedging, and humanizer pass. Use when asked to "draft the paper", "write up the results", or "write the intro".
disable-model-invocation	true
argument-hint	[section: intro | empirical-strategy | results | conclusion | abstract | full] [paper path (optional)]
allowed-tools	["Read","Grep","Glob","Write","Edit","Task"]

Draft Paper

Draft an academic economics paper (or specific section) by dispatching the Writer agent.

Input: $ARGUMENTS — section name (intro, empirical-strategy, results, conclusion, abstract, full) optionally followed by a paper path.

---

Workflow

Step 1: Context Gathering

Before dispatching the Writer:

1. Read existing paper draft in output/paper/ (if it exists)
2. Read most recent quality_reports/research_spec_*.md or quality_reports/lit_review_*.md
3. Read .claude/rules/domain-profile.md for field conventions
4. Check bibliography file for available citations
5. Scan output/*/tables/ and output/*/figures/ for generated output
6. Read quality_reports/results_summary.md if it exists (from Coder)

Step 2: Section Routing

Based on $ARGUMENTS:

• full: Draft all sections in sequence, pausing between major sections for user feedback
• intro: Draft introduction (most common request)
• empirical-strategy: Draft identification and estimation section
• results: Draft results from available output
• conclusion: Draft conclusion
• abstract: Draft abstract (must have other sections first)
• No argument: Ask user which section to draft

Step 3: Launch Writer Agent

Delegate to the writer agent via Task tool:

Prompt: Draft the [section] section for [paper].
Follow section standards:
  - Introduction: contribution paragraph within first 2 pages, effect sizes stated
  - Empirical strategy: estimating equation displayed and numbered, assumptions explicit
  - Results: every estimate with units, magnitudes not just signs
  - Conclusion: restate with effect size, limitations, implications
Apply notation protocol (Y_it, D_it, gamma_i, delta_t, epsilon_it).
Apply anti-hedging rules (ban "interestingly", "it is worth noting", etc.).
Run humanizer pass before finalizing.
Save to output/paper/sections/[section_name].tex

The Writer follows these standards:

Introduction (~1,000-1,500 words)

• Hook → research question → what we do → what we find → contribution → road map
• Contribution paragraph names specific papers being advanced
• Effect sizes with magnitudes and units

Empirical Strategy (~800-1,200 words)

• Identification assumption stated formally
• Estimating equation displayed and numbered
• Threats to identification addressed

Results (~800-1,500 words)

• Main specification with interpretation
• Effect sizes in economic terms
• Heterogeneity and robustness

Conclusion (~500-700 words)

• Restate with effect size
• Policy implications
• Limitations and future work

Step 4: Humanizer Pass

The Writer automatically applies the humanizer as a final pass:

• Strips 24 AI writing patterns across 4 categories (structural, lexical, rhetorical, formatting)
• Applies academic adaptation rules (preserve formal structure, keep technical terms)
• This is built into the Writer agent — no separate dispatch needed

Step 5: Quality Self-Check

Before presenting the draft:

• Every displayed equation is numbered (\label{eq:...})
• All \cite{} keys exist in Bibliography_base.bib
• Introduction contribution paragraph names specific papers
• Effect sizes stated with units
• No banned hedging phrases
• Notation consistent throughout
• All tables/figures referenced exist

Step 6: Present to User

Present each section for feedback. Flag:

• TBD items: Where empirical results are needed but not yet available
• VERIFY items: Citations that need user confirmation
• PLACEHOLDER items: Effect sizes awaiting final estimates

---

Principles

• This is the user's paper, not Claude's. Match their voice and style.
• Never fabricate results. Use TBD placeholders.
• Citations must be verifiable. Only cite papers you can confirm exist.
• Humanizer is automatic. Every draft gets de-AI-ified before presentation.
• LaTeX best practices. \citet for textual, \citep for parenthetical, booktabs for tables.