Build Context

• .tex -- LaTeX source (richest; extracts section structure, labels, equations, citations)
• .pdf -- PDF file (read with the Read tool; extracts prose content and structure)
• .md -- Markdown notes (author's own description of the paper, key results, terminology)

If no arguments are provided, scan the current project directory for main.tex or *.pdf files and prompt the user to select.

Workflow

Step 1: Read Source Files

1. Read each provided file
2. For .tex: extract the full document
3. For .pdf: read the PDF (use pages parameter for large files; start with first 10 pages, then continue if needed)
4. For .md: read as-is (these are usually the author's own notes)

Step 2: Extract Paper Identity

From the source material, identify:

• Title: from \title{} or document header
• Authors: from \author{} or byline
• Abstract: from \begin{abstract} or first section
• Core contribution: 1-3 sentences summarizing what the paper does that is new

Step 3: Extract Key Results

Identify the paper's main quantitative and qualitative findings:

• Primary results: the headline findings (with specific numbers if available)
• Secondary results: supporting findings
• Robustness: what robustness checks are performed
• Null results: any important non-findings

For each result, record:

• The claim as stated in the paper
• The specific numbers (magnitudes, t-statistics, confidence intervals)
• Which table/figure reports it

Step 4: Extract Terminology

Build a terminology table:

• Preferred terms: the paper's canonical names for its concepts, methods, variables
• Deprecated terms: informal or older names that should not be used
• Abbreviations: all abbreviations and their definitions

Format as:

| Preferred | Avoid | Notes |
|-----------|-------|-------|
| [correct term] | [incorrect/informal term] | [context] |

Step 5: Extract Structure

If working from .tex, record:

• Section keys: all %% BEGIN:<key> / %% END:<key> markers found
• Key labels: important \label{}s for tables, figures, equations, theorems
• Section summaries: 1-sentence summary of each section's content

Step 6: Extract Sample Description

• Date range of the sample
• Number of observations (cross-sectional units, time periods)
• Data sources
• Key filters or sample construction choices

Step 7: Extract Citation Context

Identify the paper's most important citations:

• Papers it builds directly on
• Papers it contrasts with
• Methodological references

Step 8: Write paper-context.md

Write the file to guidance/paper-context.md (creating the guidance/ directory if needed). Use this template:

# Paper Context

Generated by `/build-context` on [date]. Source files: [list].

## Identity

- **Title**: [title]
- **Authors**: [authors]
- **Core contribution**: [1-3 sentences]

## Abstract

[abstract text]

## Key Results

### Primary
1. [Result]: [magnitude] ([table/figure reference])
2. ...

### Secondary
1. ...

### Null Results
1. ...

## Terminology

| Preferred | Avoid | Notes |
|-----------|-------|-------|
| ... | ... | ... |

## Abbreviations

| Abbreviation | Definition |
|-------------|------------|
| ... | ... |

## Sample

- **Period**: [start] to [end]
- **Observations**: [count] ([unit])
- **Sources**: [data sources]
- **Key filters**: [any important sample restrictions]

## Section Structure

| Key | Section Title | Summary |
|-----|--------------|---------|
| `introduction` | Introduction | ... |
| ... | ... | ... |

## Key Labels

- Tables: [list of table labels]
- Figures: [list of figure labels]
- Equations: [list of key equation labels]

## Important Citations

- [Author (Year)]: [why it matters to this paper]
- ...

Step 9: Confirm with User

After writing, print:

1. The file path written
2. A summary of what was extracted
3. Any gaps marked [HUMAN EDIT REQUIRED] where information could not be determined from the source files

Ask the user to review and fill in any gaps.

Integration

This file is consumed by:

• /audit-section (Step 3: factual accuracy cross-reference)
• /check-consistency (Step 1: canonical values)
• /edit-section (Step 2: correct claims and framing)
• /full-paper-audit (Step 3: cross-section consistency)
• /audit-captions (Step 5: terminology compliance)
• /outline (Step 3: project-specific structural checks)
• /research (Step 4/7: relevance assessment)

Re-running

If the paper evolves, re-run /build-context to regenerate. The skill will overwrite the existing guidance/paper-context.md. The user should review the diff to ensure nothing important was lost.

Edge Cases

• Multiple papers in one project: Name the context file after the paper (e.g., guidance/paper-context-main.md). Skills that read guidance/paper-context.md will use the default; the user can specify alternatives.
• No source files found: Report what was searched and suggest the user provide a file path explicitly.
• PDF too large: Read in chunks (10 pages at a time). Summarize each chunk, then synthesize.
• Related paper (not the user's own): Generate context focused on what's relevant for reference — key findings, methodology, terminology. Mark clearly as "Reference paper context, not the project's own paper."