Write Section Skill

Write or revise a section of the thesis. This is a generalizable skill that adapts to any chapter based on section-specific templates and the established conventions from Chapters 3-5.

Arguments

• introduction — Redirects to the /write-introduction skill (specialized)
• literature [section] — Write/revise a section of Ch.2 (e.g., literature 2.1.4)
• abstract — Write the thesis abstract (~250 words)
• conclusion — Write Ch.6 (Conclusion)
• appendix A or appendix B — Write appendix content
• revise <chapter> — Revise an existing chapter (e.g., revise data, revise methodology)

If argument is introduction, invoke the /write-introduction skill instead.

Pre-Writing Protocol (ALL sections)

Before writing any section, you MUST:

1. Read the target file to understand current state (stub? outline? partial draft? complete?)
2. Read thesis/CLAUDE.md for conventions, bib keys, custom commands
3. Read CLAUDE.md (root) for project context, measure definitions, findings
4. Read .claude/reference/literature-evidence.md for concrete findings from 20+ papers — use for any section that references literature
5. Read at least 40 lines from 03_data.tex, 04_methodology.tex, and 05_results.tex to internalize the writing style
6. Read references.bib to know which citations are available

Using the Narrative Architect

For literature-heavy sections (Ch.2 subsections, Ch.1 Antecedents, Ch.6 Interpretation), dispatch the narrative-architect agent BEFORE drafting. Provide:

• Goal: What the section needs to establish
• Target section: e.g., "§2.2 — Economic Shocks and Political Behavior"
• Constraints: Length, must-cite papers, emphasis points

The architect returns a paragraph-by-paragraph outline with specific paper citations and transition strategies. Use this as the backbone for your draft.

Style Requirements (from Chapters 3-5)

Section-Specific Templates

Abstract (~250 words)

Structure:
1. Context (1-2 sentences): The phenomenon and why it matters
2. Research question (1 sentence): Clear, precise
3. Method (2-3 sentences): Data, measure, estimation strategy
4. Findings (2-3 sentences): Main results with magnitudes
5. Contribution (1-2 sentences): What this adds to the literature

Rules:

• No citations in the abstract
• No forward references
• Must stand completely alone
• Use present tense throughout
• Every sentence must be informative (no filler)

Literature Review Section (Ch.2)

Structure per section:
1. Opening sentence: What this section covers and why
2. Core papers (3-5 per subsection): Narrative synthesis, NOT paper-by-paper summary
3. Comparative passages: "While X does A, Y does B, I instead do C"
4. Forward reference: Bridge to this thesis's approach
5. Closing transition to next section

Rules:

• Synthesize, don't summarize: Group papers by finding/methodology, not chronologically
• Position the thesis: Every subsection must contain at least one passage comparing prior work to this thesis
• Forward references: Use \Cref{ch:methodology}, \Cref{ch:data}, etc.
• Discursive comparison (from style-paraphraser): "While \citet{frey2018robot} uses commuting zones, I exploit group-level variation..."
• No orphan citations: Every \cite key must exist in references.bib
• Use the evidence base: Consult .claude/reference/literature-evidence.md for concrete findings and magnitudes. Use Section 7 flow templates as starting points for the narrative-architect agent.

New Section: §2.1.4 — The Global Decline of the Labor Share

This section needs to be ADDED to Ch.2. Template:

Content:
- Karabarbounis & Neiman (2014): Global labor share decline, investment goods prices
- Autor et al. (2020): Superstar firms and labor share
- Barkai (2020): Declining labor share and rising profit share
- Connection to task displacement: labor share decline as MEASUREMENT strategy
- Forward ref to \Cref{subsec:industry_td}: "I exploit this relationship..."

This section bridges the labor economics literature to the measurement strategy in Ch.4. It explains WHY declining labor share captures automation exposure.

Expanded Section: §2.2.3 — Shift-Share Identification

Expand the existing methodological parallels section:

Additional content:
- Borusyak, Hull & Jaravel (2022): Shift-share as IV, conditions for valid inference
- Goldsmith-Pinkham, Sorkin & Swift (2020): Shares-based vs shifts-based identification
- Connection to THIS thesis's design: predetermined shares from ACS, industry-level shifts from BEA-KLEMS
- Forward ref to \Cref{sec:empirical}

Shortened Section: §2.4 — Contribution

After /write-introduction migrates the detailed contribution to Ch.1:

New structure (1-2 paragraphs only):
- Brief enumeration of 4 contributions
- "As articulated in \Cref{ch:introduction}, this thesis contributes..."
- No detailed exposition (that's now in Ch.1)

Conclusion (Ch.6)

Structure:
1. Summary of findings (2-3 paragraphs)
   - Restate RQ, method, main result
   - Key magnitudes (β, lag structure)
   - Heterogeneous effects

2. Interpretation (2-3 paragraphs)
   - What the results mean for the backlash vs insecurity debate
   - Why automation → conservative (mechanism)
   - Connection to broader populism literature

3. Policy implications (1-2 paragraphs)
   - Just transitions, retraining programs
   - Political economy of automation policy

4. Limitations (1-2 paragraphs)
   - Ecological inference caveat
   - Pseudo-panel vs true panel
   - Oster δ = 0.04 (sensitivity to unobservables)
   - External validity (US only)

5. Future research (1 paragraph)
   - Individual-level panel data
   - European comparison
   - Mechanism decomposition
   - GenAI era extension

Rules:

• No new information: Everything must reference findings from Ch.5
• Honest about limitations: Don't oversell
• Forward-looking: End on future research, not caveats
• Length: ~4-6 pages, not longer

Appendix A (Additional Tables)

Content:
- Full regression tables with all coefficients
- Alternative specifications mentioned in Ch.5
- Summary statistics by demographic group
- Industry-level task displacement values

Rules:

• Every table must have a \label{tab:app_...} and be referenced from the main text
• Use threeparttable with notes
• Match the formatting of Ch.5 tables exactly

Appendix B (Additional Figures)

Content:
- Diagnostic plots (residuals, leverage, etc.)
- Additional descriptive figures
- Sensitivity analysis plots

Rules:

• Every figure must have a \label{fig:app_...} and be referenced from main text
• Include substantive figure notes
• Check that files exist in results/figures/ before referencing

Post-Writing Protocol (ALL sections)

1. Compile

cd /Users/alessandro/Projects/Tesi/thesis && tectonic main.tex

2. Cross-Reference Check

Verify all \Cref and \cref targets exist. Tectonic will show Undefined reference warnings.

3. Citation Check

Verify all \citet and \citep keys exist in references.bib. Tectonic will show Undefined citation warnings.

4. Consistency Check

• Notation matches the notation registry (.claude/rules/notation-registry.md)
• Sign convention: higher TD = more automation exposure
• Group definition: 18 groups (Edu x Gender x Age)
• Panel sizes: national 486, state 23,814

5. Style Self-Review

Re-read the draft against the style checklist above. If the prose doesn't match Chapters 3-5, revise or dispatch to the style-paraphraser agent.

Multi-Pass Workflow

For any non-trivial section (>100 lines), follow this multi-pass approach:

Pass 1: RESEARCH
  - Read relevant chapters, papers, analysis outputs
  - Identify all citations needed
  - List all cross-references needed

Pass 2: OUTLINE
  - Write paragraph-level outline (1 sentence per paragraph)
  - Verify logical flow
  - Check that all required content is covered

Pass 3: DRAFT
  - Write full prose following section template
  - Insert all citations and cross-references
  - Mark uncertain numbers with % TODO

Pass 4: STYLE
  - Match voice/rhythm to Chapters 3-5
  - Fix sentence length, transitions, paragraph structure
  - Remove filler and passive voice

Pass 5: VERIFY
  - Compile with tectonic
  - Check all references and citations
  - Content verification against source material

For short sections (<100 lines), passes 2-3 can be combined.