swarm: Command for Claude Code

/compile_paper

Convert a SWARM paper from markdown to LaTeX (if needed), compile to PDF, and open it.

Usage

/compile_paper <slug>

Examples:

• /compile_paper kernel_market_v2
• /compile_paper collusion_governance

Arguments

• slug: Paper identifier. Resolves to docs/papers/<slug>.md (source) and docs/papers/<slug>.tex (output).

Behavior

Step 1: Locate source

Look for files in this order:

1. docs/papers/<slug>.tex — if it exists and is newer than the .md, skip conversion
2. docs/papers/<slug>.md — convert to LaTeX using the template

Step 2: Convert markdown to LaTeX (if needed)

If the .tex file does not exist or the .md is newer:

1. Read docs/papers/template.tex for the standard preamble
2. Read docs/papers/<slug>.md for content
3. Convert markdown to LaTeX:
   • # Title → \title{}
   • ## Section → \section{}
   • ### Subsection → \subsection{}
   • **bold** → \textbf{}
   • *italic* → \textit{}
   • `code` → \texttt{}
   • $math$ → pass through
   • Tables → \begin{table}[H] with booktabs
   • Lists → \begin{itemize} / \begin{enumerate}
   • Code blocks → \begin{verbatim}
   • % → \%, _ in prose → \_
4. Write docs/papers/<slug>.tex

Use the template preamble exactly. Replace %TITLE%, %AUTHOR%, %ABSTRACT%, and %BODY% placeholders with converted content.

For %AUTHOR%, resolve in order:

1. $SWARM_AUTHOR environment variable
2. git config user.name
3. Ask the user (do not guess from OS username)

Step 2b: Auto-copy figures

Before compiling, ensure referenced figures are available:

1. Scan the .tex for \includegraphics{...} paths
2. For each referenced figure not found in docs/papers/figures/:
   • Search runs/*/plots/ for a file with the same name
   • If found, copy it to docs/papers/figures/<slug>/
3. This prevents "File not found" errors during tectonic compilation

Step 3: Compile

cd docs/papers/ && tectonic <slug>.tex